{ "schema_version": "1.0", "generated": "2026-10-15", "generated_by": "Inside the Machine data agent", "methodology": { "memory_formula": "min_memory_gb = weights_gb_4bit × 1.15 + 2. weights_gb_4bit = Q4_K_M file (Ollama, else HF GGUF) → MLX 4-bit (HF) → Ollama MLX/NVFP4 tag → Ollama default tag → params_total_b × 0.6 (flagged weights_estimated). +15% ≈ KV cache at ~8–16K context; +2GB runtime/process overhead. Long contexts (128K+) need much more.", "tier_rule": "A model fits a tier when min_memory_gb ≤ 0.75 × tier_gb. 0.75 approximates the default share of unified memory macOS lets the GPU wire (can be raised with `sudo sysctl iogpu.wired_limit_mb=`), and leaves room for OS/apps on PCs. For discrete GPUs, treat tier_gb as VRAM (CPU offload is possible but slow).", "file_sizes": "Copied as listed on Ollama tag pages (GB, rounded by Ollama) or summed from Hugging Face repo file sizes (bytes/1e9). Retrieved 2026-10-15.", "benchmarks": "independent = Artificial Analysis (current index version as displayed 2026-10-15; not comparable to older AA index numbers); vendor-reported = parsed from the model card table on Hugging Face (column matched to this model; verify before quoting); crowd-preference = LMArena Text leaderboard Elo (2026-10-15).", "gap": "Derived only: AA points behind the top closed model, and Elo behind LMArena #1. Not the site's Gap Score (that needs Mac Studio tests)." }, "sources": [ { "name": "Ollama library", "url": "https://ollama.com/library" }, { "name": "Hugging Face Hub API", "url": "https://huggingface.co/api/models" }, { "name": "Artificial Analysis leaderboard", "url": "https://artificialanalysis.ai/leaderboards/models" }, { "name": "LMArena Text leaderboard", "url": "https://lmarena.ai/leaderboard/text" } ], "frontier_reference": { "artificial_analysis": [ { "name": "Claude Opus 5.5 (Max, Default Fallback)", "vendor": "Anthropic", "open_weights": false, "aa_intelligence_index": 57.6, "gpqa_pct": null, "hle_pct": 61.4, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ 4, 20 ], "url": "https://artificialanalysis.ai/models/claude-opus-5-5", "as_of": "2026-10-15" }, { "name": "Muse Wonder Full (Max)", "vendor": "Meta", "open_weights": false, "aa_intelligence_index": 55.8, "gpqa_pct": null, "hle_pct": 53.6, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ null, null ], "url": "https://artificialanalysis.ai/models/muse-wonder-full", "as_of": "2026-10-15" }, { "name": "GPT-6 Astra (Max)", "vendor": "OpenAI", "open_weights": false, "aa_intelligence_index": 52.7, "gpqa_pct": 96.1, "hle_pct": 54.7, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ 10, 50 ], "url": "https://artificialanalysis.ai/models/gpt-6-astra", "as_of": "2026-10-15" }, { "name": "Gemini 4 Argon (High)", "vendor": "Google", "open_weights": false, "aa_intelligence_index": 52.6, "gpqa_pct": null, "hle_pct": 57.1, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ 2, 10 ], "url": "https://artificialanalysis.ai/models/gemini-4-argon", "as_of": "2026-10-15" }, { "name": "Grok 4.7 (Xhigh)", "vendor": "SpaceXAI", "open_weights": false, "aa_intelligence_index": 46.4, "gpqa_pct": null, "hle_pct": 43.1, "terminal_bench_hard_pct": null, "context_tokens": 500000, "price_per_1m_in_out_usd": [ 2, 6 ], "url": "https://artificialanalysis.ai/models/grok-4-7", "as_of": "2026-10-15" }, { "name": "Qwen3.8 Max (0902)", "vendor": "Alibaba", "open_weights": false, "aa_intelligence_index": 45.4, "gpqa_pct": 92.8, "hle_pct": 43.1, "terminal_bench_hard_pct": null, "context_tokens": 983616, "price_per_1m_in_out_usd": [ 2, 6 ], "url": "https://artificialanalysis.ai/models/qwen3-8-max", "as_of": "2026-10-15" }, { "name": "Step 5 Preview", "vendor": "StepFun", "open_weights": false, "aa_intelligence_index": 43.7, "gpqa_pct": null, "hle_pct": 46.5, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ 1, 2.7 ], "url": "https://artificialanalysis.ai/models/step-5", "as_of": "2026-10-15" }, { "name": "Ling 3.1 Flash", "vendor": "InclusionAI", "open_weights": false, "aa_intelligence_index": 41.1, "gpqa_pct": null, "hle_pct": 39.4, "terminal_bench_hard_pct": null, "context_tokens": 1000000, "price_per_1m_in_out_usd": [ 0.3, 0.9 ], "url": "https://artificialanalysis.ai/models/ling-3-1-flash", "as_of": "2026-10-15" } ], "lmarena": [ { "arena_name": "gemini-4-argon-high", "vendor": "Google", "elo": 1524, "ci": 9, "votes": 4904, "rank": 1, "preliminary": true, "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15" }, { "arena_name": "muse-wonder-max", "vendor": "Meta", "elo": 1509, "ci": 11, "votes": 3247, "rank": 2, "preliminary": false, "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15" }, { "arena_name": "claude-opus-5.5-high", "vendor": "Anthropic", "elo": 1508, "ci": 7, "votes": 8300, "rank": 3, "preliminary": false, "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15" }, { "arena_name": "gpt-6.1-sol-max", "vendor": "OpenAI", "elo": 1484, "ci": 8, "votes": 6320, "rank": 22, "preliminary": false, "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15" }, { "arena_name": "qwen3.8-max", "vendor": "Alibaba (proprietary)", "elo": 1482, "ci": 5, "votes": 27766, "rank": 24, "preliminary": false, "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15" } ], "best_open_weight_aa": [ { "id": "mimo-v2.6-pro", "aa": 46.3 }, { "id": "glm-5.3", "aa": 44.8 }, { "id": "kimi-k3", "aa": 43.6 }, { "id": "glm-5.3-flash", "aa": 41.8 }, { "id": "qwen3.8-2.4t-a95b", "aa": 39.9 } ] }, "ram_tiers": [ { "tier_gb": 8, "usable_gb_assumed": 6.0, "top_picks": [ { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "aa_index_best": 13.1, "lmarena_elo": null, "min_memory_gb": 5.8, "headroom_gb": 0.2, "fits_q8": false, "weights_estimated": false }, { "id": "minicpm5-2b", "name": "MiniCPM5 2B", "aa_index_best": 12.5, "lmarena_elo": null, "min_memory_gb": 3.9, "headroom_gb": 2.1, "fits_q8": true, "weights_estimated": false }, { "id": "granite-4.2-3b", "name": "Granite 4.2 3B", "aa_index_best": 9.1, "lmarena_elo": 1290, "min_memory_gb": 4.5, "headroom_gb": 1.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-2b", "name": "Qwen3.5 2B", "aa_index_best": 6.9, "lmarena_elo": null, "min_memory_gb": 4.2, "headroom_gb": 1.8, "fits_q8": true, "weights_estimated": false }, { "id": "phi-4-mini", "name": "Phi-4-mini 3.8B", "aa_index_best": 6.3, "lmarena_elo": null, "min_memory_gb": 4.9, "headroom_gb": 1.1, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-0.8b", "name": "Qwen3.5 0.8B", "aa_index_best": 6.1, "lmarena_elo": null, "min_memory_gb": 2.6, "headroom_gb": 3.4, "fits_q8": true, "weights_estimated": false } ], "best_coding": null, "best_vision": { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "aa_index_best": 13.1, "lmarena_elo": null, "min_memory_gb": 5.8, "headroom_gb": 0.2, "fits_q8": false, "weights_estimated": false }, "best_agentic": null, "unlocked_at_this_tier": [ { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "aa_index_best": 13.1, "lmarena_elo": null, "min_memory_gb": 5.8, "headroom_gb": 0.2, "fits_q8": false, "weights_estimated": false }, { "id": "minicpm5-2b", "name": "MiniCPM5 2B", "aa_index_best": 12.5, "lmarena_elo": null, "min_memory_gb": 3.9, "headroom_gb": 2.1, "fits_q8": true, "weights_estimated": false }, { "id": "granite-4.2-3b", "name": "Granite 4.2 3B", "aa_index_best": 9.1, "lmarena_elo": 1290, "min_memory_gb": 4.5, "headroom_gb": 1.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-2b", "name": "Qwen3.5 2B", "aa_index_best": 6.9, "lmarena_elo": null, "min_memory_gb": 4.2, "headroom_gb": 1.8, "fits_q8": true, "weights_estimated": false }, { "id": "phi-4-mini", "name": "Phi-4-mini 3.8B", "aa_index_best": 6.3, "lmarena_elo": null, "min_memory_gb": 4.9, "headroom_gb": 1.1, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "embeddinggemma-2", "fit_count": 19, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 16, "usable_gb_assumed": 12.0, "top_picks": [ { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 3.4, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-12b", "name": "Gemma 4 12B", "aa_index_best": 14.2, "lmarena_elo": null, "min_memory_gb": 11.2, "headroom_gb": 0.8, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-9b", "name": "Qwen3.5 9B", "aa_index_best": 13.3, "lmarena_elo": null, "min_memory_gb": 9.6, "headroom_gb": 2.4, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "aa_index_best": 13.1, "lmarena_elo": null, "min_memory_gb": 5.8, "headroom_gb": 6.2, "fits_q8": true, "weights_estimated": false }, { "id": "minicpm5-2b", "name": "MiniCPM5 2B", "aa_index_best": 12.5, "lmarena_elo": null, "min_memory_gb": 3.9, "headroom_gb": 8.1, "fits_q8": true, "weights_estimated": false }, { "id": "granite-4.2-8b", "name": "Granite 4.2 8B", "aa_index_best": 11.1, "lmarena_elo": 1290, "min_memory_gb": 8.1, "headroom_gb": 3.9, "fits_q8": false, "weights_estimated": false } ], "best_coding": { "id": "ornith-1.5-9b", "name": "Ornith 1.5 9B", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 8.8, "headroom_gb": 3.2, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "gemma-4-12b", "name": "Gemma 4 12B", "aa_index_best": 14.2, "lmarena_elo": null, "min_memory_gb": 11.2, "headroom_gb": 0.8, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "mellum2.1-12b-a2.5b", "name": "Mellum2.1 12B-A2.5B Thinking", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 11.3, "headroom_gb": 0.7, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 3.4, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-12b", "name": "Gemma 4 12B", "aa_index_best": 14.2, "lmarena_elo": null, "min_memory_gb": 11.2, "headroom_gb": 0.8, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-9b", "name": "Qwen3.5 9B", "aa_index_best": 13.3, "lmarena_elo": null, "min_memory_gb": 9.6, "headroom_gb": 2.4, "fits_q8": false, "weights_estimated": false }, { "id": "granite-4.2-8b", "name": "Granite 4.2 8B", "aa_index_best": 11.1, "lmarena_elo": 1290, "min_memory_gb": 8.1, "headroom_gb": 3.9, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-e4b", "name": "Gemma 4 E4B", "aa_index_best": 8.9, "lmarena_elo": null, "min_memory_gb": 9.6, "headroom_gb": 2.4, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 38, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 24, "usable_gb_assumed": 18.0, "top_picks": [ { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 9.4, "fits_q8": true, "weights_estimated": false }, { "id": "gemma-4-12b", "name": "Gemma 4 12B", "aa_index_best": 14.2, "lmarena_elo": null, "min_memory_gb": 11.2, "headroom_gb": 6.8, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-9b", "name": "Qwen3.5 9B", "aa_index_best": 13.3, "lmarena_elo": null, "min_memory_gb": 9.6, "headroom_gb": 8.4, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "aa_index_best": 13.1, "lmarena_elo": null, "min_memory_gb": 5.8, "headroom_gb": 12.2, "fits_q8": true, "weights_estimated": false }, { "id": "minicpm5-2b", "name": "MiniCPM5 2B", "aa_index_best": 12.5, "lmarena_elo": null, "min_memory_gb": 3.9, "headroom_gb": 14.1, "fits_q8": true, "weights_estimated": false }, { "id": "granite-4.2-8b", "name": "Granite 4.2 8B", "aa_index_best": 11.1, "lmarena_elo": 1290, "min_memory_gb": 8.1, "headroom_gb": 9.9, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "nouscoder-14b", "name": "NousCoder 14B", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 12.3, "headroom_gb": 5.7, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "gemma-4-12b", "name": "Gemma 4 12B", "aa_index_best": 14.2, "lmarena_elo": null, "min_memory_gb": 11.2, "headroom_gb": 6.8, "fits_q8": true, "weights_estimated": false }, "best_agentic": { "id": "mellum2.1-12b-a2.5b", "name": "Mellum2.1 12B-A2.5B Thinking", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 11.3, "headroom_gb": 6.7, "fits_q8": true, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "gpt-oss-20b", "name": "gpt-oss-20b", "aa_index_best": 10.0, "lmarena_elo": 1317, "min_memory_gb": 15.4, "headroom_gb": 2.6, "fits_q8": false, "weights_estimated": false }, { "id": "ministral-3-14b", "name": "Ministral 3 14B", "aa_index_best": 6.0, "lmarena_elo": null, "min_memory_gb": 12.5, "headroom_gb": 5.5, "fits_q8": false, "weights_estimated": false }, { "id": "phi-4-14b", "name": "Phi-4 14B", "aa_index_best": 5.9, "lmarena_elo": 1256, "min_memory_gb": 12.5, "headroom_gb": 5.5, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-3n-e4b", "name": "Gemma 3n E4B", "aa_index_best": 4.8, "lmarena_elo": 1318, "min_memory_gb": 13.3, "headroom_gb": 4.7, "fits_q8": true, "weights_estimated": false }, { "id": "phi-4-reasoning-plus", "name": "Phi-4-reasoning-plus 14B", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 14.6, "headroom_gb": 3.4, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 45, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 32, "usable_gb_assumed": 24.0, "top_picks": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 2.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 2.5, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 15.4, "fits_q8": true, "weights_estimated": false }, { "id": "muse-glimmer-30b", "name": "Muse Glimmer 30B", "aa_index_best": 17.5, "lmarena_elo": 1425, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-26b-a4b", "name": "Gemma 4 26B-A4B", "aa_index_best": 16.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 2.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 2.5, "fits_q8": false, "weights_estimated": false }, { "id": "muse-glimmer-30b", "name": "Muse Glimmer 30B", "aa_index_best": 17.5, "lmarena_elo": 1425, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-26b-a4b", "name": "Gemma 4 26B-A4B", "aa_index_best": 16.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 1.3, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 62, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 48, "usable_gb_assumed": 36.0, "top_picks": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 13.3, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 7.9, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 14.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 14.5, "fits_q8": false, "weights_estimated": false }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 30.7, "headroom_gb": 5.3, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 27.4, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 13.3, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 13.3, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 13.3, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 7.9, "fits_q8": false, "weights_estimated": false }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 30.7, "headroom_gb": 5.3, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-35b-a3b", "name": "Qwen3.5 35B-A3B", "aa_index_best": 19.3, "lmarena_elo": 1394, "min_memory_gb": 27.3, "headroom_gb": 8.7, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.6-35b-a3b", "name": "Qwen3.6 35B-A3B", "aa_index_best": 18.2, "lmarena_elo": null, "min_memory_gb": 29.6, "headroom_gb": 6.4, "fits_q8": false, "weights_estimated": false }, { "id": "gemma-4-31b", "name": "Gemma 4 31B", "aa_index_best": 14.7, "lmarena_elo": 1452, "min_memory_gb": 25.0, "headroom_gb": 11.0, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 76, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 64, "usable_gb_assumed": 48.0, "top_picks": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 25.3, "fits_q8": true, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 19.9, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 26.5, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 26.5, "fits_q8": true, "weights_estimated": false }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 30.7, "headroom_gb": 17.3, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 39.4, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 25.3, "fits_q8": true, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 25.3, "fits_q8": true, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 25.3, "fits_q8": true, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "kimi-linear-48b-a3b", "name": "Kimi Linear 48B-A3B", "aa_index_best": 7.3, "lmarena_elo": null, "min_memory_gb": 36.6, "headroom_gb": 11.4, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 77, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 96, "usable_gb_assumed": 72.0, "top_picks": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 49.3, "fits_q8": true, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 43.9, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 50.5, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 50.5, "fits_q8": true, "weights_estimated": false }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 30.7, "headroom_gb": 41.3, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 63.4, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 49.3, "fits_q8": true, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 49.3, "fits_q8": true, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 49.3, "fits_q8": true, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "qwen3-next-80b-a3b", "name": "Qwen3-Next 80B-A3B", "aa_index_best": 11.2, "lmarena_elo": 1399, "min_memory_gb": 59.5, "headroom_gb": 12.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3-coder-next", "name": "Qwen3-Coder-Next (80B-A3B)", "aa_index_best": 9.2, "lmarena_elo": null, "min_memory_gb": 61.8, "headroom_gb": 10.2, "fits_q8": false, "weights_estimated": false }, { "id": "hermes-4-70b", "name": "Hermes 4 70B (Llama 3.1)", "aa_index_best": 7.9, "lmarena_elo": null, "min_memory_gb": 50.9, "headroom_gb": 21.1, "fits_q8": false, "weights_estimated": false }, { "id": "llama-3.3-70b", "name": "Llama 3.3 70B", "aa_index_best": 7.7, "lmarena_elo": 1318, "min_memory_gb": 51.4, "headroom_gb": 20.6, "fits_q8": false, "weights_estimated": false }, { "id": "kolibri-1", "name": "Kolibri 1", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 56.6, "headroom_gb": 15.4, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 83, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 128, "usable_gb_assumed": 96.0, "top_picks": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 73.3, "fits_q8": true, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 67.9, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 74.5, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 74.5, "fits_q8": true, "weights_estimated": false }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 30.7, "headroom_gb": 65.3, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "aa_index_best": 20.6, "lmarena_elo": null, "min_memory_gb": 8.6, "headroom_gb": 87.4, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 73.3, "fits_q8": true, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 73.3, "fits_q8": true, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 73.3, "fits_q8": true, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "qwen3.5-122b-a10b", "name": "Qwen3.5 122B-A10B", "aa_index_best": 17.7, "lmarena_elo": 1416, "min_memory_gb": 95.1, "headroom_gb": 0.9, "fits_q8": false, "weights_estimated": false }, { "id": "mistral-medium-3.5-128b", "name": "Mistral Medium 3.5 128B", "aa_index_best": 14.2, "lmarena_elo": 1426, "min_memory_gb": 94.0, "headroom_gb": 2.0, "fits_q8": false, "weights_estimated": false }, { "id": "gpt-oss-120b", "name": "gpt-oss-120b", "aa_index_best": 11.6, "lmarena_elo": 1352, "min_memory_gb": 74.2, "headroom_gb": 21.8, "fits_q8": false, "weights_estimated": false }, { "id": "mistral-small-4-119b", "name": "Mistral Small 4 119B", "aa_index_best": 11.3, "lmarena_elo": null, "min_memory_gb": 85.0, "headroom_gb": 11.0, "fits_q8": false, "weights_estimated": false }, { "id": "glm-4.5-air", "name": "GLM-4.5-Air 106B-A12B", "aa_index_best": 11.1, "lmarena_elo": 1373, "min_memory_gb": 85.9, "headroom_gb": 10.1, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 90, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 192, "usable_gb_assumed": 144.0, "top_picks": [ { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 4.0, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 121.3, "fits_q8": true, "weights_estimated": false }, { "id": "nemotron-3.5-super-120b", "name": "Nemotron 3.5 Super VL 120B-A12B", "aa_index_best": 26.5, "lmarena_elo": 1416, "min_memory_gb": 105.5, "headroom_gb": 38.5, "fits_q8": false, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 115.9, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 122.5, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "aa_index_best": 21.4, "lmarena_elo": null, "min_memory_gb": 21.5, "headroom_gb": 122.5, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 4.0, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 4.0, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 4.0, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 4.0, "fits_q8": false, "weights_estimated": false }, { "id": "nemotron-3.5-super-120b", "name": "Nemotron 3.5 Super VL 120B-A12B", "aa_index_best": 26.5, "lmarena_elo": 1416, "min_memory_gb": 105.5, "headroom_gb": 38.5, "fits_q8": false, "weights_estimated": false }, { "id": "nemotron-3-super-120b", "name": "Nemotron 3 Super 120B-A12B", "aa_index_best": 12.8, "lmarena_elo": 1360, "min_memory_gb": 102.0, "headroom_gb": 42.0, "fits_q8": false, "weights_estimated": false }, { "id": "laguna-s-2.1", "name": "Laguna S 2.1 118B-A8B", "aa_index_best": null, "lmarena_elo": null, "min_memory_gb": 112.4, "headroom_gb": 31.6, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 94, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 256, "usable_gb_assumed": 192.0, "top_picks": [ { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 52.0, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 169.3, "fits_q8": true, "weights_estimated": false }, { "id": "nemotron-3.5-super-120b", "name": "Nemotron 3.5 Super VL 120B-A12B", "aa_index_best": 26.5, "lmarena_elo": 1416, "min_memory_gb": 105.5, "headroom_gb": 86.5, "fits_q8": true, "weights_estimated": false }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "aa_index_best": 25.3, "lmarena_elo": null, "min_memory_gb": 28.1, "headroom_gb": 163.9, "fits_q8": true, "weights_estimated": false }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "aa_index_best": 22.9, "lmarena_elo": 1409, "min_memory_gb": 21.5, "headroom_gb": 170.5, "fits_q8": true, "weights_estimated": false }, { "id": "minimax-m2.7", "name": "MiniMax M2.7", "aa_index_best": 22.8, "lmarena_elo": 1415, "min_memory_gb": 161.1, "headroom_gb": 30.9, "fits_q8": false, "weights_estimated": false } ], "best_coding": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 52.0, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 52.0, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 52.0, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "minimax-m2.7", "name": "MiniMax M2.7", "aa_index_best": 22.8, "lmarena_elo": 1415, "min_memory_gb": 161.1, "headroom_gb": 30.9, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 95, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." }, { "tier_gb": 512, "usable_gb_assumed": 384.0, "top_picks": [ { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "aa_index_best": 41.8, "lmarena_elo": 1474, "min_memory_gb": 233.0, "headroom_gb": 151.0, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "aa_index_best": 39.8, "lmarena_elo": null, "min_memory_gb": 140.0, "headroom_gb": 244.0, "fits_q8": true, "weights_estimated": false }, { "id": "mimo-v2.6-flash", "name": "MiMo-V2.6-Flash", "aa_index_best": 37.9, "lmarena_elo": 1451, "min_memory_gb": 216.4, "headroom_gb": 167.6, "fits_q8": false, "weights_estimated": true }, { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "aa_index_best": 33.7, "lmarena_elo": 1438, "min_memory_gb": 22.7, "headroom_gb": 361.3, "fits_q8": true, "weights_estimated": false }, { "id": "minimax-m3", "name": "MiniMax M3", "aa_index_best": 29.2, "lmarena_elo": 1441, "min_memory_gb": 302.5, "headroom_gb": 81.5, "fits_q8": false, "weights_estimated": false }, { "id": "nemotron-3.5-super-120b", "name": "Nemotron 3.5 Super VL 120B-A12B", "aa_index_best": 26.5, "lmarena_elo": 1416, "min_memory_gb": 105.5, "headroom_gb": 278.5, "fits_q8": true, "weights_estimated": false } ], "best_coding": { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "aa_index_best": 41.8, "lmarena_elo": 1474, "min_memory_gb": 233.0, "headroom_gb": 151.0, "fits_q8": false, "weights_estimated": false }, "best_vision": { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "aa_index_best": 41.8, "lmarena_elo": 1474, "min_memory_gb": 233.0, "headroom_gb": 151.0, "fits_q8": false, "weights_estimated": false }, "best_agentic": { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "aa_index_best": 41.8, "lmarena_elo": 1474, "min_memory_gb": 233.0, "headroom_gb": 151.0, "fits_q8": false, "weights_estimated": false }, "unlocked_at_this_tier": [ { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "aa_index_best": 41.8, "lmarena_elo": 1474, "min_memory_gb": 233.0, "headroom_gb": 151.0, "fits_q8": false, "weights_estimated": false }, { "id": "mimo-v2.6-flash", "name": "MiMo-V2.6-Flash", "aa_index_best": 37.9, "lmarena_elo": 1451, "min_memory_gb": 216.4, "headroom_gb": 167.6, "fits_q8": false, "weights_estimated": true }, { "id": "minimax-m3", "name": "MiniMax M3", "aa_index_best": 29.2, "lmarena_elo": 1441, "min_memory_gb": 302.5, "headroom_gb": 81.5, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3.5-397b-a17b", "name": "Qwen3.5 397B-A17B", "aa_index_best": 21.4, "lmarena_elo": 1442, "min_memory_gb": 282.7, "headroom_gb": 101.3, "fits_q8": false, "weights_estimated": false }, { "id": "qwen3-coder-480b-a35b", "name": "Qwen3-Coder 480B-A35B", "aa_index_best": 11.9, "lmarena_elo": 1388, "min_memory_gb": 335.5, "headroom_gb": 48.5, "fits_q8": false, "weights_estimated": false } ], "embedding_pick": "qwen3-embedding-8b", "fit_count": 102, "ranking_basis": "Highest Artificial Analysis Intelligence Index (best listed variant) among models whose min_memory_gb ≤ usable_gb; ties broken by LMArena Elo. Unscored models are excluded from top_picks. unlocked_at_this_tier = models that need more than the previous tier allowed (includes unscored, ranked by AA then Elo)." } ], "models": [ { "id": "qwen3.8-27b", "name": "Qwen3.8 27B", "family": "Qwen3.8", "vendor": "Alibaba Qwen", "categories": [ "general", "coding", "vision", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-08-05", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 27.78, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.8" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 18.0, "tag": "qwen3.8:27b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 30.0, "tag": "qwen3.8:27b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 56.0, "tag": "qwen3.8:27b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 18.0, "tag": "qwen3.8:27b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 32.0, "tag": "qwen3.8:27b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 18.0, "tag": "qwen3.8:27b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.05, "repo": "mlx-community/Qwen3.8-27B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.8-27B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 18.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 22.7, "min_memory_gb_q8": 36.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8 27B (Xhigh)", "value": 33.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8 27B (Xhigh)", "value": 90.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8 27B (Xhigh)", "value": 33.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8 27B (Xhigh)", "value": 79.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8 27B (Xhigh)", "value": 46.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8 27B (Xhigh)", "value": 82.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.8 27B (Xhigh)", "value": 76.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8 27B (Medium)", "value": 27.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8 27B (Medium)", "value": 84.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8 27B (Medium)", "value": 14.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8 27B (Medium)", "value": 65.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8 27B (Medium)", "value": 39.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8 27B (Medium)", "value": 79.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.8 27B (Medium)", "value": 74.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-medium", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8 27B (Low)", "value": 26.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8 27B (Low)", "value": 84.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8 27B (Low)", "value": 14.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8 27B (Low)", "value": 67.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8 27B (Low)", "value": 40.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8 27B (Low)", "value": 77.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.8 27B (Low)", "value": 73.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 20.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 81.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 12.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 49.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 36.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 69.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.8 27B (Non-reasoning)", "value": 69.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Agentic terminal coding Terminal Bench 2.1 (Terminus)", "value": 73.0, "unit": "%", "column": "Qwen3.8-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "as_of": "2026-08-14", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "Scientific reasoning GPQA Diamond", "value": 89.2, "unit": "%", "column": "Qwen3.8-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "as_of": "2026-08-14", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "Competitive coding LiveCodeBench v6", "value": 90.3, "unit": "%", "column": "Qwen3.8-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "as_of": "2026-08-14", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3.8-27b", "value": 1438, "unit": "elo", "ci": 5, "votes": 23294, "rank": 100, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 33.7, "lmarena_elo": 1438 }, "popularity": { "ollama_pulls": "3.9M", "hf_downloads_30d": 7363968, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.8-27B", "ollama": "https://ollama.com/library/qwen3.8", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.8-27B-4bit" }, "run": { "ollama": "ollama run qwen3.8:27b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.8-27B-4bit --prompt 'hello'" }, "description": "Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks.", "notes": [], "gap": { "aa_points_behind_frontier": 23.9, "aa_pct_of_frontier": 58.5, "lmarena_elo_behind_frontier": 86, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.8-flash-next", "name": "Qwen3.8-Flash-Next (125B-A6B)", "family": "Qwen3.8", "vendor": "Alibaba Qwen", "categories": [ "general", "coding", "vision", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-08-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 180.0, "active_b": 6, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"125B with 6B activated, plus 51B n-gram embedding and 4B MTP\" (https://huggingface.co/Qwen/Qwen3.8-Flash-Next)" }, "license": { "id": "other", "name": "qwen-community-1.0", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.8-flash-next" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 120.0, "tag": "qwen3.8-flash-next:125b-a6b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8-flash-next/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 189.0, "tag": "qwen3.8-flash-next:125b-a6b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8-flash-next/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 355.0, "tag": "qwen3.8-flash-next:125b-a6b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8-flash-next/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 105.0, "tag": "qwen3.8-flash-next:125b-a6b-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.8-flash-next/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 111.52, "repo": "mlx-community/Qwen3.8-Flash-Next-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.8-Flash-Next-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 120.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 140.0, "min_memory_gb_q8": 219.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8-Flash-Next", "value": 39.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8-Flash-Next", "value": 92.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8-Flash-Next", "value": 38.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8-Flash-Next", "value": 86.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8-Flash-Next", "value": 50.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8-Flash-Next", "value": 79.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.8-Flash-Next", "value": 79.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-flash-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Scientific reasoning GPQA Diamond", "value": 91.7, "unit": "%", "column": "Qwen3.8-Flash-Next", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "as_of": "2026-08-27", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "Competitive coding LiveCodeBench v6", "value": 91.9, "unit": "%", "column": "Qwen3.8-Flash-Next", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "as_of": "2026-08-27", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 39.8, "lmarena_elo": null }, "popularity": { "ollama_pulls": "189.2K", "hf_downloads_30d": 2290500, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "ollama": "https://ollama.com/library/qwen3.8-flash-next", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.8-Flash-Next-4bit" }, "run": { "ollama": "ollama run qwen3.8-flash-next:125b-a6b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.8-Flash-Next-4bit --prompt 'hello'" }, "description": "This experimental preview of the architecture that will underpin Qwen4.", "notes": [ "Experimental preview of the Qwen4 architecture (per Ollama). HF total 180B = 125B MoE + 51B n-gram embedding + 4B MTP." ], "gap": { "aa_points_behind_frontier": 17.8, "aa_pct_of_frontier": 69.1, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.8-2.4t-a95b", "name": "Qwen3.8 2.4T-A95B", "family": "Qwen3.8", "vendor": "Alibaba Qwen", "categories": [ "general", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-08-08", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 2446.18, "active_b": 95.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "qwen3.8-max", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q8_0": { "value_gb": 2600.25, "repo": "unsloth/Qwen3.8-2.4T-A95B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Qwen3.8-2.4T-A95B-GGUF", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 1467.71, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 1689.9, "min_memory_gb_q8": 2992.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.8 2.4T A95B", "value": 39.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.8 2.4T A95B", "value": 93.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.8 2.4T A95B", "value": 42.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.8 2.4T A95B", "value": 82.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.8 2.4T A95B", "value": 54.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.8 2.4T A95B", "value": 80.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 39.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 32502, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 17.7, "aa_pct_of_frontier": 69.3, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.6-27b", "name": "Qwen3.6 27B", "family": "Qwen3.6", "vendor": "Alibaba Qwen", "categories": [ "general", "coding", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-04-21", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 27.78, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.6-27B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.6" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 17.0, "tag": "qwen3.6:27b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 30.0, "tag": "qwen3.6:27b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 56.0, "tag": "qwen3.6:27b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 19.0, "tag": "qwen3.6:27b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 30.0, "tag": "qwen3.6:27b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "qwen3.6:27b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.05, "repo": "mlx-community/Qwen3.6-27B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.6-27B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 17.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 21.5, "min_memory_gb_q8": 36.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.6 27B (Reasoning)", "value": 21.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.6 27B (Reasoning)", "value": 84.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.6 27B (Reasoning)", "value": 23.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.6 27B (Reasoning)", "value": 34.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.6 27B (Reasoning)", "value": 60.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.6 27B (Reasoning)", "value": 94.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Qwen3.6 27B (Reasoning)", "value": 42.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.6 27B (Reasoning)", "value": 67.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.6 27B (Reasoning)", "value": 77.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.6 27B (Reasoning)", "value": 74.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 19.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 82.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 15.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 21.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 51.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 93.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 45.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 66.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.6 27B (Non-reasoning)", "value": 71.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 77.2, "unit": "%", "column": "Qwen3.6-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.0", "value": 59.3, "unit": "%", "column": "Qwen3.6-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 86.2, "unit": "%", "column": "Qwen3.6-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 87.8, "unit": "%", "column": "Qwen3.6-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 83.9, "unit": "%", "column": "Qwen3.6-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 21.4, "lmarena_elo": null }, "popularity": { "ollama_pulls": "7.1M", "hf_downloads_30d": 2288557, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.6-27B", "ollama": "https://ollama.com/library/qwen3.6", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.6-27B-4bit" }, "run": { "ollama": "ollama run qwen3.6:27b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.6-27B-4bit --prompt 'hello'" }, "description": "Qwen3.6 delivers substantial upgrades in agentic coding and thinking preservation than previous Qwen models.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 36.2, "aa_pct_of_frontier": 37.2, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.6-35b-a3b", "name": "Qwen3.6 35B-A3B", "family": "Qwen3.6", "vendor": "Alibaba Qwen", "categories": [ "general", "coding", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-04-15", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 35.95, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.6" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 24.0, "tag": "qwen3.6:35b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 39.0, "tag": "qwen3.6:35b-a3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 72.0, "tag": "qwen3.6:35b-a3b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 24.0, "tag": "qwen3.6:35b-a3b-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 38.0, "tag": "qwen3.6:35b-a3b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 24.0, "tag": "qwen3.6:35b-a3b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.6/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 20.4, "repo": "mlx-community/Qwen3.6-35B-A3B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.6-35B-A3B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 24.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 29.6, "min_memory_gb_q8": 46.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 18.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 84.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 22.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 34.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 44.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 95.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 36.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 64.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 71.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.6 35B A3B (Reasoning)", "value": 75.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 15.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 81.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 13.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 25.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 41.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 85.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 36.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 64.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.6 35B A3B (Non-reasoning)", "value": 71.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 73.4, "unit": "%", "column": "Qwen3.6-35BA3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.0", "value": 51.5, "unit": "%", "column": "Qwen3.6-35BA3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 85.2, "unit": "%", "column": "Qwen3.6-35BA3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 86.0, "unit": "%", "column": "Qwen3.6-35BA3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 80.4, "unit": "%", "column": "Qwen3.6-35BA3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 18.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "7.1M", "hf_downloads_30d": 3695301, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "ollama": "https://ollama.com/library/qwen3.6", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.6-35B-A3B-4bit" }, "run": { "ollama": "ollama run qwen3.6:35b-a3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.6-35B-A3B-4bit --prompt 'hello'" }, "description": "Qwen3.6 delivers substantial upgrades in agentic coding and thinking preservation than previous Qwen models.", "notes": [], "gap": { "aa_points_behind_frontier": 39.4, "aa_pct_of_frontier": 31.6, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-0.8b", "name": "Qwen3.5 0.8B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "edge", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.87, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q8_0": { "value_gb": 1.0, "tag": "qwen3.5:0.8b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 1.8, "tag": "qwen3.5:0.8b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 1.2, "tag": "qwen3.5:0.8b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 1.2, "tag": "qwen3.5:0.8b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 1.3, "tag": "qwen3.5:0.8b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 0.53, "repo": "unsloth/Qwen3.5-0.8B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Qwen3.5-0.8B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 0.63, "repo": "mlx-community/Qwen3.5-0.8B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-0.8B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.53, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 2.6, "min_memory_gb_q8": 3.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 6.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 11.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 1.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 47.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 21.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 9.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 0.8B (Reasoning)", "value": 25.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 5.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 23.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 5.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 0.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 65.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 21.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 8.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 0.8B (Non-reasoning)", "value": 25.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 29.7, "unit": "%", "column": "Qwen3.5-0.8B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 11.9, "unit": "%", "column": "Qwen3.5-0.8B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B", "as_of": "2026-03-02", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 6.1, "lmarena_elo": null }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 2880351, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-0.8B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-0.8B-4bit" }, "run": { "ollama": "ollama run qwen3.5:0.8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-0.8B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [], "gap": { "aa_points_behind_frontier": 51.5, "aa_pct_of_frontier": 10.6, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-2b", "name": "Qwen3.5 2B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "edge", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 2.27, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-2B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 1.9, "tag": "qwen3.5:2b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 2.7, "tag": "qwen3.5:2b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 4.6, "tag": "qwen3.5:2b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 3.1, "tag": "qwen3.5:2b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 3.1, "tag": "qwen3.5:2b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 3.1, "tag": "qwen3.5:2b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 1.72, "repo": "mlx-community/Qwen3.5-2B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-2B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 1.9, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 4.2, "min_memory_gb_q8": 5.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 2B (Reasoning)", "value": 6.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 2B (Reasoning)", "value": 45.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 2B (Reasoning)", "value": 2.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 2B (Reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 2B (Reasoning)", "value": 3.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 2B (Reasoning)", "value": 69.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 2B (Reasoning)", "value": 31.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 2B (Reasoning)", "value": 21.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 2B (Reasoning)", "value": 43.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 6.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 43.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 5.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 81.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 29.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 14.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 2B (Non-reasoning)", "value": 42.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 55.3, "unit": "%", "column": "Qwen3.5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-2B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 51.6, "unit": "%", "column": "Qwen3.5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-2B", "as_of": "2026-03-02", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 6.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 2814601, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-2B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-2B-4bit" }, "run": { "ollama": "ollama run qwen3.5:2b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-2B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [], "gap": { "aa_points_behind_frontier": 50.7, "aa_pct_of_frontier": 12.0, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-4b", "name": "Qwen3.5 4B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "edge", "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-27", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 4.66, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-4B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 3.3, "tag": "qwen3.5:4b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 5.2, "tag": "qwen3.5:4b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 9.3, "tag": "qwen3.5:4b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 4.0, "tag": "qwen3.5:4b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 5.6, "tag": "qwen3.5:4b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 4.0, "tag": "qwen3.5:4b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 3.03, "repo": "mlx-community/Qwen3.5-4B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-4B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 3.3, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 5.8, "min_memory_gb_q8": 8.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 4B (Reasoning)", "value": 13.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 4B (Reasoning)", "value": 77.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 4B (Reasoning)", "value": 9.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 4B (Reasoning)", "value": 18.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 4B (Reasoning)", "value": 25.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 4B (Reasoning)", "value": 92.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 4B (Reasoning)", "value": 52.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 4B (Reasoning)", "value": 63.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 4B (Reasoning)", "value": 65.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 10.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 71.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 8.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 11.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 21.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 87.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 33.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 34.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 4B (Non-reasoning)", "value": 62.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 79.1, "unit": "%", "column": "Qwen3.5-4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-4B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 76.2, "unit": "%", "column": "Qwen3.5-4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-4B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 55.8, "unit": "%", "column": "Qwen3.5-4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-4B", "as_of": "2026-03-02", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 13.1, "lmarena_elo": null }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 7944995, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-4B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-4B-4bit" }, "run": { "ollama": "ollama run qwen3.5:4b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-4B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [], "gap": { "aa_points_behind_frontier": 44.5, "aa_pct_of_frontier": 22.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-9b", "name": "Qwen3.5 9B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-27", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 9.65, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-9B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 6.6, "tag": "qwen3.5:9b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 10.0, "tag": "qwen3.5:9b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 19.0, "tag": "qwen3.5:9b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 8.9, "tag": "qwen3.5:9b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 11.0, "tag": "qwen3.5:9b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 7.6, "tag": "qwen3.5:9b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 5.95, "repo": "mlx-community/Qwen3.5-9B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-9B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 6.6, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 9.6, "min_memory_gb_q8": 13.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 9B (Reasoning)", "value": 11.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 9B (Reasoning)", "value": 80.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 9B (Reasoning)", "value": 14.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 9B (Reasoning)", "value": 24.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 9B (Reasoning)", "value": 29.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 9B (Reasoning)", "value": 86.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.5 9B (Reasoning)", "value": 29.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 9B (Reasoning)", "value": 66.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 9B (Reasoning)", "value": 70.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 9B (Reasoning)", "value": 69.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 13.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 78.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 9.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 18.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 21.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 85.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 37.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 46.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 9B (Non-reasoning)", "value": 66.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 82.5, "unit": "%", "column": "Qwen3.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 81.7, "unit": "%", "column": "Qwen3.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "as_of": "2026-03-02", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 65.6, "unit": "%", "column": "Qwen3.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "as_of": "2026-03-02", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 13.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 8591266, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-9B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-9B-4bit" }, "run": { "ollama": "ollama run qwen3.5:9b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-9B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [], "gap": { "aa_points_behind_frontier": 44.3, "aa_pct_of_frontier": 23.1, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-27b", "name": "Qwen3.5 27B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "general", "vision" ], "status": "superseded", "open_weights": true, "release_date": "2026-02-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 27.78, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-27B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 17.0, "tag": "qwen3.5:27b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 30.0, "tag": "qwen3.5:27b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 56.0, "tag": "qwen3.5:27b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 20.0, "tag": "qwen3.5:27b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 30.0, "tag": "qwen3.5:27b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 20.0, "tag": "qwen3.5:27b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.05, "repo": "mlx-community/Qwen3.5-27B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-27B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 17.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 21.5, "min_memory_gb_q8": 36.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 27B (Reasoning)", "value": 22.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 27B (Reasoning)", "value": 85.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 27B (Reasoning)", "value": 23.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 27B (Reasoning)", "value": 32.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 27B (Reasoning)", "value": 93.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.5 27B (Reasoning)", "value": 75.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 27B (Reasoning)", "value": 77.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 27B (Reasoning)", "value": 75.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 19.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 84.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 13.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 31.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 87.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 46.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 64.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 27B (Non-reasoning)", "value": 70.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 86.1, "unit": "%", "column": "Qwen3.5-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 85.5, "unit": "%", "column": "Qwen3.5-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 72.4, "unit": "%", "column": "Qwen3.5-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2", "value": 41.6, "unit": "%", "column": "Qwen3.5-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 80.7, "unit": "%", "column": "Qwen3.5-27B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3.5-27b", "value": 1409, "unit": "elo", "ci": 4, "votes": 28671, "rank": 153, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 22.9, "lmarena_elo": 1409 }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 1840434, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-27B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-27B-4bit" }, "run": { "ollama": "ollama run qwen3.5:27b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-27B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 34.7, "aa_pct_of_frontier": 39.8, "lmarena_elo_behind_frontier": 115, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-35b-a3b", "name": "Qwen3.5 35B-A3B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "general", "vision" ], "status": "superseded", "open_weights": true, "release_date": "2026-02-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 35.95, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 22.0, "tag": "qwen3.5:35b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 39.0, "tag": "qwen3.5:35b-a3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 72.0, "tag": "qwen3.5:35b-a3b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 22.0, "tag": "qwen3.5:35b-a3b-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 37.0, "tag": "qwen3.5:35b-a3b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 22.0, "tag": "qwen3.5:35b-a3b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 20.39, "repo": "mlx-community/Qwen3.5-35B-A3B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-35B-A3B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 22.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 27.3, "min_memory_gb_q8": 46.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 19.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 84.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 21.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 26.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 89.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 72.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 72.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 35B A3B (Reasoning)", "value": 72.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 15.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 81.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 13.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 10.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 40.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 86.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 44.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 63.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 35B A3B (Non-reasoning)", "value": 69.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 85.3, "unit": "%", "column": "Qwen3.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 84.2, "unit": "%", "column": "Qwen3.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 69.2, "unit": "%", "column": "Qwen3.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2", "value": 40.5, "unit": "%", "column": "Qwen3.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 74.6, "unit": "%", "column": "Qwen3.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3.5-35b-a3b", "value": 1394, "unit": "elo", "ci": 4, "votes": 30457, "rank": 169, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 19.3, "lmarena_elo": 1394 }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 1484223, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-35B-A3B-4bit" }, "run": { "ollama": "ollama run qwen3.5:35b-a3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-35B-A3B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 38.3, "aa_pct_of_frontier": 33.5, "lmarena_elo_behind_frontier": 130, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-122b-a10b", "name": "Qwen3.5 122B-A10B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 125.09, "active_b": 10.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 81.0, "tag": "qwen3.5:122b-a10b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 81.0, "tag": "qwen3.5:122b-a10b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q8_0": { "value_gb": 129.87, "repo": "unsloth/Qwen3.5-122B-A10B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Qwen3.5-122B-A10B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 69.59, "repo": "mlx-community/Qwen3.5-122B-A10B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-122B-A10B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 81.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 95.1, "min_memory_gb_q8": 151.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 15.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 85.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 25.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 31.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 47.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 93.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 39.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 75.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 76.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 122B A10B (Reasoning)", "value": 75.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 17.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 82.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 15.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 29.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 47.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 84.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 50.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 61.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 122B A10B (Non-reasoning)", "value": 70.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 86.7, "unit": "%", "column": "Qwen3.5-122B-A10B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 86.6, "unit": "%", "column": "Qwen3.5-122B-A10B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 72.0, "unit": "%", "column": "Qwen3.5-122B-A10B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2", "value": 49.4, "unit": "%", "column": "Qwen3.5-122B-A10B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 78.9, "unit": "%", "column": "Qwen3.5-122B-A10B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3.5-122b-a10b", "value": 1416, "unit": "elo", "ci": 4, "votes": 29732, "rank": 136, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 17.7, "lmarena_elo": 1416 }, "popularity": { "ollama_pulls": "22.4M", "hf_downloads_30d": 677777, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "ollama": "https://ollama.com/library/qwen3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-122B-A10B-4bit" }, "run": { "ollama": "ollama run qwen3.5:122b-a10b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-122B-A10B-4bit --prompt 'hello'" }, "description": "Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.", "notes": [], "gap": { "aa_points_behind_frontier": 39.9, "aa_pct_of_frontier": 30.7, "lmarena_elo_behind_frontier": 108, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3.5-397b-a17b", "name": "Qwen3.5 397B-A17B", "family": "Qwen3.5", "vendor": "Alibaba Qwen", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-02-16", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 403.4, "active_b": 17.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 244.09, "repo": "unsloth/Qwen3.5-397B-A17B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Qwen3.5-397B-A17B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 421.51, "repo": "unsloth/Qwen3.5-397B-A17B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Qwen3.5-397B-A17B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 223.86, "repo": "mlx-community/Qwen3.5-397B-A17B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3.5-397B-A17B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 244.09, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 282.7, "min_memory_gb_q8": 486.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 18.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 89.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 29.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 40.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 51.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 95.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 44.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 78.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 77.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 397B A17B (Reasoning)", "value": 77.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 21.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 86.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 19.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 35.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 83.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 51.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 64.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Qwen3.5 397B A17B (Non-reasoning)", "value": 52.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 87.8, "unit": "%", "column": "Qwen3.5-397B-A17B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 88.4, "unit": "%", "column": "Qwen3.5-397B-A17B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 83.6, "unit": "%", "column": "Qwen3.5-397B-A17B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 76.4, "unit": "%", "column": "Qwen3.5-397B-A17B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2", "value": 52.5, "unit": "%", "column": "Qwen3.5-397B-A17B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "as_of": "2026-04-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3.5-397b-a17b", "value": 1442, "unit": "elo", "ci": 3, "votes": 93567, "rank": 94, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 21.4, "lmarena_elo": 1442 }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 562196, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3.5-397B-A17B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/Qwen3.5-397B-A17B-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 36.2, "aa_pct_of_frontier": 37.2, "lmarena_elo_behind_frontier": 82, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-coder-30b-a3b", "name": "Qwen3-Coder 30B-A3B", "family": "Qwen3-Coder", "vendor": "Alibaba Qwen", "categories": [ "coding" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-07-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 30.53, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-coder" }, "inputs": null, "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "qwen3-coder:30b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 32.0, "tag": "qwen3-coder:30b-a3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 61.0, "tag": "qwen3-coder:30b-a3b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 17.18, "repo": "mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 38.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 9.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 51.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 15.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 34.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 32.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 Coder 30B A3B Instruct", "value": 32.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 9.6, "lmarena_elo": null }, "popularity": { "ollama_pulls": "10M", "hf_downloads_30d": 422336, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "ollama": "https://ollama.com/library/qwen3-coder", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit" }, "run": { "ollama": "ollama run qwen3-coder:30b-a3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bit --prompt 'hello'" }, "description": "Alibaba's performant long context models for agentic and coding tasks.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 48.0, "aa_pct_of_frontier": 16.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-coder-next", "name": "Qwen3-Coder-Next (80B-A3B)", "family": "Qwen3-Coder", "vendor": "Alibaba Qwen", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-01-30", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 79.67, "active_b": 3, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"only 3B activated parameters (80B total parameters)\" (https://huggingface.co/Qwen/Qwen3-Coder-Next)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-coder-next" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 52.0, "tag": "qwen3-coder-next:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder-next/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 85.0, "tag": "qwen3-coder-next:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder-next/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 52.0, "tag": "qwen3-coder-next:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder-next/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 44.84, "repo": "mlx-community/Qwen3-Coder-Next-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-Coder-Next-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 52.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 61.8, "min_memory_gb_q8": 99.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 Coder Next", "value": 9.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 Coder Next", "value": 73.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 Coder Next", "value": 10.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 Coder Next", "value": 18.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3 Coder Next", "value": 38.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 Coder Next", "value": 79.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Qwen3 Coder Next", "value": 36.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3 Coder Next", "value": 35.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 Coder Next", "value": 47.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-next", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 9.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "2.1M", "hf_downloads_30d": 588526, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "ollama": "https://ollama.com/library/qwen3-coder-next", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-Coder-Next-4bit" }, "run": { "ollama": "ollama run qwen3-coder-next", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-Coder-Next-4bit --prompt 'hello'" }, "description": "Qwen3-Coder-Next is a coding-focused language model from Alibaba's Qwen team, optimized for agentic coding workflows and local development.", "notes": [], "gap": { "aa_points_behind_frontier": 48.4, "aa_pct_of_frontier": 16.0, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-coder-480b-a35b", "name": "Qwen3-Coder 480B-A35B", "family": "Qwen3-Coder", "vendor": "Alibaba Qwen", "categories": [ "coding" ], "status": "legacy", "open_weights": true, "release_date": "2025-07-22", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 480.15, "active_b": 35.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-coder" }, "inputs": null, "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 290.0, "tag": "qwen3-coder:480b-a35b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 510.0, "tag": "qwen3-coder:480b-a35b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 960.0, "tag": "qwen3-coder:480b-a35b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-coder/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 270.09, "repo": "mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 290.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 335.5, "min_memory_gb_q8": 588.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 11.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 61.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 18.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 43.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 40.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 Coder 480B A35B Instruct", "value": 45.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "qwen3-coder-480b-a35b-instruct", "value": 1388, "unit": "elo", "ci": 5, "votes": 25195, "rank": 180, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 11.9, "lmarena_elo": 1388 }, "popularity": { "ollama_pulls": "10M", "hf_downloads_30d": 39504, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct", "ollama": "https://ollama.com/library/qwen3-coder", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit" }, "run": { "ollama": "ollama run qwen3-coder:480b-a35b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit --prompt 'hello'" }, "description": "Alibaba's performant long context models for agentic and coding tasks.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 45.7, "aa_pct_of_frontier": 20.7, "lmarena_elo_behind_frontier": 136, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-next-80b-a3b", "name": "Qwen3-Next 80B-A3B", "family": "Qwen3-Next", "vendor": "Alibaba Qwen", "categories": [ "general" ], "status": "legacy", "open_weights": true, "release_date": "2025-09-09", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 81.32, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-next" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 50.0, "tag": "qwen3-next:80b-a3b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-next/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 85.0, "tag": "qwen3-next:80b-a3b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-next/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 159.0, "tag": "qwen3-next:80b-a3b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-next/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 44.84, "repo": "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 50.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 59.5, "min_memory_gb_q8": 99.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 Next 80B A3B Instruct", "value": 9.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 Next 80B A3B Instruct", "value": 73.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 Next 80B A3B Instruct", "value": 7.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 Next 80B A3B Instruct", "value": 7.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 Next 80B A3B Instruct", "value": 21.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3 Next 80B A3B Instruct", "value": 39.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 Next 80B A3B Instruct", "value": 52.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 11.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 75.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 12.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 9.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 6.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 41.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 60.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 Next 80B A3B (Reasoning)", "value": 63.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 80.6, "unit": "%", "column": "Qwen3-Next-80B-A3B-Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "as_of": "2025-09-17", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 72.9, "unit": "%", "column": "Qwen3-Next-80B-A3B-Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "as_of": "2025-09-17", "retrieved": "2026-10-15" }, { "benchmark": "AIME 2025", "variant": "AIME25", "value": 69.5, "unit": "%", "column": "Qwen3-Next-80B-A3B-Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "as_of": "2025-09-17", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6 (25.02-25.05)", "value": 56.6, "unit": "%", "column": "Qwen3-Next-80B-A3B-Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "as_of": "2025-09-17", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "qwen3-next-80b-a3b-instruct", "value": 1399, "unit": "elo", "ci": 5, "votes": 22760, "rank": 164, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 11.2, "lmarena_elo": 1399 }, "popularity": { "ollama_pulls": "601.3K", "hf_downloads_30d": 202355, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "ollama": "https://ollama.com/library/qwen3-next", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit" }, "run": { "ollama": "ollama run qwen3-next:80b-a3b-instruct", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit --prompt 'hello'" }, "description": "The first installment in the Qwen3-Next series with strong performance in terms of both parameter efficiency and inference speed.", "notes": [], "gap": { "aa_points_behind_frontier": 46.4, "aa_pct_of_frontier": 19.4, "lmarena_elo_behind_frontier": 125, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-30b-a3b-2507", "name": "Qwen3 30B-A3B (2507)", "family": "Qwen3", "vendor": "Alibaba Qwen", "categories": [ "general" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-07-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 30.53, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "qwen3:30b-a3b-instruct-2507-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 32.0, "tag": "qwen3:30b-a3b-instruct-2507-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 61.0, "tag": "qwen3:30b-a3b-instruct-2507-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 17.18, "repo": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 38.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 7.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 65.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 6.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 6.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 10.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 33.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 30B A3B 2507 Instruct", "value": 26.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 9.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 70.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 10.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 1.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 28.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 33.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 50.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 30B A3B 2507 (Reasoning)", "value": 61.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "qwen3-30b-a3b-instruct-2507", "value": 1382, "unit": "elo", "ci": 5, "votes": 23322, "rank": 188, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 9.8, "lmarena_elo": 1382 }, "popularity": { "ollama_pulls": "40.2M", "hf_downloads_30d": 1051915, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "ollama": "https://ollama.com/library/qwen3", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit" }, "run": { "ollama": "ollama run qwen3:30b-a3b-instruct-2507", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit --prompt 'hello'" }, "description": "Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 47.8, "aa_pct_of_frontier": 17.0, "lmarena_elo_behind_frontier": 142, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-8b", "name": "Qwen3 8B", "family": "Qwen3", "vendor": "Alibaba Qwen", "categories": [ "general" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-04-27", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.19, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 40960, "native_source": "HF config.json", "ollama_default": 40960, "ollama_source": "https://ollama.com/library/qwen3" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 5.2, "tag": "qwen3:8b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 8.9, "tag": "qwen3:8b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "qwen3:8b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 5.2, "tag": "qwen3:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 4.61, "repo": "mlx-community/Qwen3-8B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-8B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.2, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.0, "min_memory_gb_q8": 12.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 8B (Non-reasoning)", "value": 6.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 8B (Non-reasoning)", "value": 45.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 8B (Non-reasoning)", "value": 1.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 8B (Non-reasoning)", "value": 2.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 8B (Non-reasoning)", "value": 24.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 8B (Non-reasoning)", "value": 28.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 8B (Non-reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3 8B (Reasoning)", "value": 5.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 8B (Reasoning)", "value": 58.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 8B (Reasoning)", "value": 3.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 8B (Reasoning)", "value": 2.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Qwen3 8B (Reasoning)", "value": 2.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 8B (Reasoning)", "value": 27.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Qwen3 8B (Reasoning)", "value": 27.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 8B (Reasoning)", "value": 33.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 8B (Reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 6.0, "lmarena_elo": null }, "popularity": { "ollama_pulls": "40.2M", "hf_downloads_30d": 9578964, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-8B", "ollama": "https://ollama.com/library/qwen3", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-8B-4bit" }, "run": { "ollama": "ollama run qwen3:8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-8B-4bit --prompt 'hello'" }, "description": "Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 51.6, "aa_pct_of_frontier": 10.4, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-vl-8b", "name": "Qwen3-VL 8B", "family": "Qwen3-VL", "vendor": "Alibaba Qwen", "categories": [ "vision" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-10-11", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.77, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-vl" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 6.1, "tag": "qwen3-vl:8b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.8, "tag": "qwen3-vl:8b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 18.0, "tag": "qwen3-vl:8b-instruct-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 6.1, "tag": "qwen3-vl:8b-instruct", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 5.76, "repo": "mlx-community/Qwen3-VL-8B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-VL-8B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 6.1, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 9.0, "min_memory_gb_q8": 13.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 VL 8B Instruct", "value": 7.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 VL 8B Instruct", "value": 42.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 VL 8B Instruct", "value": 2.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 VL 8B Instruct", "value": 2.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 VL 8B Instruct", "value": 29.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 VL 8B Instruct", "value": 32.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 VL 8B Instruct", "value": 16.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3 VL 8B Instruct", "value": 47.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3 VL 8B (Reasoning)", "value": 8.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 VL 8B (Reasoning)", "value": 57.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 VL 8B (Reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 VL 8B (Reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 VL 8B (Reasoning)", "value": 22.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 VL 8B (Reasoning)", "value": 39.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Qwen3 VL 8B (Reasoning)", "value": 33.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3 VL 8B (Reasoning)", "value": 56.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 8.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "6.7M", "hf_downloads_30d": 6999638, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct", "ollama": "https://ollama.com/library/qwen3-vl", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-VL-8B-Instruct-4bit" }, "run": { "ollama": "ollama run qwen3-vl:8b-instruct", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-VL-8B-Instruct-4bit --prompt 'hello'" }, "description": "The most powerful vision-language model in the Qwen model family to date. ", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 49.4, "aa_pct_of_frontier": 14.2, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-vl-30b-a3b", "name": "Qwen3-VL 30B-A3B", "family": "Qwen3-VL", "vendor": "Alibaba Qwen", "categories": [ "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-09-30", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 31.07, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/qwen3-vl" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 20.0, "tag": "qwen3-vl:30b-a3b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 34.0, "tag": "qwen3-vl:30b-a3b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 62.0, "tag": "qwen3-vl:30b-a3b-instruct-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 20.0, "tag": "qwen3-vl:30b-a3b-instruct", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-vl/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.25, "repo": "mlx-community/Qwen3-VL-30B-A3B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Qwen3-VL-30B-A3B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 20.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 25.0, "min_memory_gb_q8": 41.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Qwen3 VL 30B A3B Instruct", "value": 7.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 VL 30B A3B Instruct", "value": 69.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 VL 30B A3B Instruct", "value": 6.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 VL 30B A3B Instruct", "value": 6.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 VL 30B A3B Instruct", "value": 19.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 VL 30B A3B Instruct", "value": 33.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3 VL 30B A3B Instruct", "value": 62.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 9.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 72.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 8.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 19.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 45.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Qwen3 VL 30B A3B (Reasoning)", "value": 61.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 9.5, "lmarena_elo": null }, "popularity": { "ollama_pulls": "6.7M", "hf_downloads_30d": 273635, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct", "ollama": "https://ollama.com/library/qwen3-vl", "mlx_4bit": "https://huggingface.co/mlx-community/Qwen3-VL-30B-A3B-Instruct-4bit" }, "run": { "ollama": "ollama run qwen3-vl:30b-a3b-instruct", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Qwen3-VL-30B-A3B-Instruct-4bit --prompt 'hello'" }, "description": "The most powerful vision-language model in the Qwen model family to date. ", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 48.1, "aa_pct_of_frontier": 16.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-embedding-0.6b", "name": "Qwen3-Embedding 0.6B", "family": "Qwen3-Embedding", "vendor": "Alibaba Qwen", "categories": [ "embedding" ], "status": "current", "open_weights": true, "release_date": "2025-06-03", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.6, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 32768, "native_source": "HF config.json", "ollama_default": 32768, "ollama_source": "https://ollama.com/library/qwen3-embedding" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "q8_0": { "value_gb": 0.639, "tag": "qwen3-embedding:0.6b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 1.2, "tag": "qwen3-embedding:0.6b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.639, "tag": "qwen3-embedding:0.6b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 0.639, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 2.7, "min_memory_gb_q8": 2.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "4.3M", "hf_downloads_30d": 8605583, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "ollama": "https://ollama.com/library/qwen3-embedding", "mlx_4bit": null }, "run": { "ollama": "ollama run qwen3-embedding:0.6b", "ollama_cloud_only": false, "mlx": null }, "description": "Building upon the foundational models of the Qwen3 series, Qwen3 Embedding provides a comprehensive range of text embeddings models in various sizes", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "qwen3-embedding-8b", "name": "Qwen3-Embedding 8B", "family": "Qwen3-Embedding", "vendor": "Alibaba Qwen", "categories": [ "embedding" ], "status": "current", "open_weights": true, "release_date": "2025-06-03", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 7.57, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Qwen/Qwen3-Embedding-8B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 40960, "native_source": "HF config.json", "ollama_default": 40960, "ollama_source": "https://ollama.com/library/qwen3-embedding" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "q4_k_m": { "value_gb": 4.7, "tag": "qwen3-embedding:8b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 8.0, "tag": "qwen3-embedding:8b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 15.0, "tag": "qwen3-embedding:8b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 4.7, "tag": "qwen3-embedding:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/qwen3-embedding/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 4.7, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.4, "min_memory_gb_q8": 11.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "4.3M", "hf_downloads_30d": 2036891, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/Qwen/Qwen3-Embedding-8B", "ollama": "https://ollama.com/library/qwen3-embedding", "mlx_4bit": null }, "run": { "ollama": "ollama run qwen3-embedding:8b", "ollama_cloud_only": false, "mlx": null }, "description": "Building upon the foundational models of the Qwen3 series, Qwen3 Embedding provides a comprehensive range of text embeddings models in various sizes", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-4-e2b", "name": "Gemma 4 E2B", "family": "Gemma 4", "vendor": "Google", "categories": [ "edge", "vision", "audio" ], "status": "current", "open_weights": true, "release_date": "2026-03-02", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 5.12, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gemma4" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking", "audio" ], "files": { "q4_k_m": { "value_gb": 4.6, "tag": "gemma4:e2b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 6.0, "tag": "gemma4:e2b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 10.0, "tag": "gemma4:e2b-it-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 7.5, "tag": "gemma4:e2b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 8.2, "tag": "gemma4:e2b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 7.5, "tag": "gemma4:e2b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 3.55, "repo": "mlx-community/gemma-4-e2b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-4-e2b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 4.6, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.3, "min_memory_gb_q8": 8.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 E2B (Reasoning)", "value": 7.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 E2B (Reasoning)", "value": 43.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 E2B (Reasoning)", "value": 4.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 E2B (Reasoning)", "value": 3.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 E2B (Reasoning)", "value": 0.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 E2B (Reasoning)", "value": 20.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 E2B (Reasoning)", "value": 38.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 E2B (Reasoning)", "value": 16.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 E2B (Reasoning)", "value": 44.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 6.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 40.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 4.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 2.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 22.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 33.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 16.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 E2B (Non-reasoning)", "value": 41.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 44.0, "unit": "%", "column": "Gemma 4 E2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-E2B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 43.4, "unit": "%", "column": "Gemma 4 E2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-E2B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 7.8, "lmarena_elo": null }, "popularity": { "ollama_pulls": "27.1M", "hf_downloads_30d": 2974727, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-4-E2B-it", "ollama": "https://ollama.com/library/gemma4", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-4-e2b-it-4bit" }, "run": { "ollama": "ollama run gemma4:e2b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-4-e2b-it-4bit --prompt 'hello'" }, "description": "Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.", "notes": [ "\"E2B\" = ~2B effective parameters; HF safetensors total (5.1B) includes per-layer embeddings and audio/vision encoders." ], "gap": { "aa_points_behind_frontier": 49.8, "aa_pct_of_frontier": 13.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-4-e4b", "name": "Gemma 4 E4B", "family": "Gemma 4", "vendor": "Google", "categories": [ "edge", "vision", "audio" ], "status": "current", "open_weights": true, "release_date": "2026-03-02", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.0, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gemma4" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking", "audio" ], "files": { "q4_k_m": { "value_gb": 6.6, "tag": "gemma4:e4b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.0, "tag": "gemma4:e4b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "gemma4:e4b-it-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 9.5, "tag": "gemma4:e4b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 12.0, "tag": "gemma4:e4b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 9.5, "tag": "gemma4:e4b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 5.15, "repo": "mlx-community/gemma-4-e4b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-4-e4b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 6.6, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 9.6, "min_memory_gb_q8": 12.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 E4B (Reasoning)", "value": 8.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 E4B (Reasoning)", "value": 57.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 E4B (Reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 E4B (Reasoning)", "value": 8.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 E4B (Reasoning)", "value": 1.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 E4B (Reasoning)", "value": 20.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Gemma 4 E4B (Reasoning)", "value": 24.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 E4B (Reasoning)", "value": 44.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 E4B (Reasoning)", "value": 32.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 E4B (Reasoning)", "value": 51.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 7.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 54.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 4.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 7.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 26.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 40.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 24.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 E4B (Non-reasoning)", "value": 51.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 52.0, "unit": "%", "column": "Gemma 4 E4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-E4B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 58.6, "unit": "%", "column": "Gemma 4 E4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-E4B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 8.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": "27.1M", "hf_downloads_30d": 4193163, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-4-E4B-it", "ollama": "https://ollama.com/library/gemma4", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-4-e4b-it-4bit" }, "run": { "ollama": "ollama run gemma4:e4b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-4-e4b-it-4bit --prompt 'hello'" }, "description": "Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.", "notes": [ "\"E4B\" = ~4B effective parameters; HF total (8.0B) includes per-layer embeddings and encoders." ], "gap": { "aa_points_behind_frontier": 48.7, "aa_pct_of_frontier": 15.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-4-12b", "name": "Gemma 4 12B", "family": "Gemma 4", "vendor": "Google", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-05-23", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 11.96, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/gemma4" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking", "audio" ], "files": { "q4_k_m": { "value_gb": 8.0, "tag": "gemma4:12b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 13.0, "tag": "gemma4:12b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 24.0, "tag": "gemma4:12b-it-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 7.7, "tag": "gemma4:12b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 13.0, "tag": "gemma4:12b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 8.0, "tag": "gemma4:12b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 6.74, "repo": "mlx-community/gemma-4-12B-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-4-12B-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 8.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 11.2, "min_memory_gb_q8": 16.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 12B (Reasoning)", "value": 14.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 12B (Reasoning)", "value": 75.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 12B (Reasoning)", "value": 15.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 12B (Reasoning)", "value": 18.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 12B (Reasoning)", "value": 27.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 12B (Reasoning)", "value": 36.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 12B (Reasoning)", "value": 73.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 12B (Reasoning)", "value": 63.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 12B (Reasoning)", "value": 69.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 12B (Non-reasoning)", "value": 9.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 12B (Non-reasoning)", "value": 66.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 12B (Non-reasoning)", "value": 6.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 12B (Non-reasoning)", "value": 11.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 12B (Non-reasoning)", "value": 31.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 12B (Non-reasoning)", "value": 45.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 12B (Non-reasoning)", "value": 35.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 12B (Non-reasoning)", "value": 62.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 72.0, "unit": "%", "column": "Gemma 4 12B Unified", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-12B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 78.8, "unit": "%", "column": "Gemma 4 12B Unified", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-12B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 14.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "27.1M", "hf_downloads_30d": 1656357, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-4-12B-it", "ollama": "https://ollama.com/library/gemma4", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-4-12B-it-4bit" }, "run": { "ollama": "ollama run gemma4:12b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-4-12B-it-4bit --prompt 'hello'" }, "description": "Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.", "notes": [], "gap": { "aa_points_behind_frontier": 43.4, "aa_pct_of_frontier": 24.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-4-26b-a4b", "name": "Gemma 4 26B-A4B", "family": "Gemma 4", "vendor": "Google", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-03-11", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 25.81, "active_b": 4, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"26B A4B ... activating a 4B subset\" (https://huggingface.co/google/gemma-4-26B-A4B-it)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/gemma4" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking", "audio" ], "files": { "q4_k_m": { "value_gb": 18.0, "tag": "gemma4:26b-a4b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 28.0, "tag": "gemma4:26b-a4b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 52.0, "tag": "gemma4:26b-a4b-it-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 18.0, "tag": "gemma4:26b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 29.0, "tag": "gemma4:26b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "gemma4:26b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 15.34, "repo": "mlx-community/gemma-4-26b-a4b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-4-26b-a4b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 18.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 22.7, "min_memory_gb_q8": 34.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 16.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 79.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 19.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 13.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 39.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 43.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 40.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 72.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 65.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 26B A4B (Reasoning)", "value": 69.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 13.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 71.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 11.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 25.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 40.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 45.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 42.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 26B A4B (Non-reasoning)", "value": 66.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 77.1, "unit": "%", "column": "Gemma 4 26B A4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 82.3, "unit": "%", "column": "Gemma 4 26B A4B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "gemma-4-26b-a4b", "value": 1438, "unit": "elo", "ci": 7, "votes": 6094, "rank": 101, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 16.7, "lmarena_elo": 1438 }, "popularity": { "ollama_pulls": "27.1M", "hf_downloads_30d": 11170734, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-4-26B-A4B-it", "ollama": "https://ollama.com/library/gemma4", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-4-26b-a4b-it-4bit" }, "run": { "ollama": "ollama run gemma4:26b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-4-26b-a4b-it-4bit --prompt 'hello'" }, "description": "Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.", "notes": [], "gap": { "aa_points_behind_frontier": 40.9, "aa_pct_of_frontier": 29.0, "lmarena_elo_behind_frontier": 86, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-4-31b", "name": "Gemma 4 31B", "family": "Gemma 4", "vendor": "Google", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-03-11", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 31.27, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://ai.google.dev/gemma/docs/gemma_4_license", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/gemma4" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking", "audio" ], "files": { "q4_k_m": { "value_gb": 20.0, "tag": "gemma4:31b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 34.0, "tag": "gemma4:31b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 63.0, "tag": "gemma4:31b-it-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 19.0, "tag": "gemma4:31b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 34.0, "tag": "gemma4:31b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 20.0, "tag": "gemma4:31b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.41, "repo": "mlx-community/gemma-4-31b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-4-31b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 20.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 25.0, "min_memory_gb_q8": 41.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 31B (Reasoning)", "value": 14.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 31B (Reasoning)", "value": 85.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 31B (Reasoning)", "value": 23.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 31B (Reasoning)", "value": 36.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 31B (Reasoning)", "value": 43.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 31B (Reasoning)", "value": 59.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Gemma 4 31B (Reasoning)", "value": 45.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 31B (Reasoning)", "value": 75.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 31B (Reasoning)", "value": 69.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 31B (Reasoning)", "value": 73.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Gemma 4 31B (Non-reasoning)", "value": 13.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Gemma 4 31B (Non-reasoning)", "value": 76.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 4 31B (Non-reasoning)", "value": 11.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 4 31B (Non-reasoning)", "value": 30.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 4 31B (Non-reasoning)", "value": 29.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 4 31B (Non-reasoning)", "value": 65.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Gemma 4 31B (Non-reasoning)", "value": 53.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 4 31B (Non-reasoning)", "value": 46.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Gemma 4 31B (Non-reasoning)", "value": 70.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 80.0, "unit": "%", "column": "Gemma 4 31B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-31B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 84.3, "unit": "%", "column": "Gemma 4 31B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/gemma-4-31B-it", "as_of": "2026-07-20", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "gemma-4-31b", "value": 1452, "unit": "elo", "ci": 7, "votes": 6147, "rank": 76, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 14.7, "lmarena_elo": 1452 }, "popularity": { "ollama_pulls": "27.1M", "hf_downloads_30d": 9180042, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-4-31B-it", "ollama": "https://ollama.com/library/gemma4", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-4-31b-it-4bit" }, "run": { "ollama": "ollama run gemma4:31b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-4-31b-it-4bit --prompt 'hello'" }, "description": "Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.", "notes": [], "gap": { "aa_points_behind_frontier": 42.9, "aa_pct_of_frontier": 25.5, "lmarena_elo_behind_frontier": 72, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-3-27b", "name": "Gemma 3 27B", "family": "Gemma 3", "vendor": "Google", "categories": [ "general", "vision" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-03-01", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 27.43, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "gemma", "name": null, "url": "https://huggingface.co/google/gemma-3-27b-it", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gemma3" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 17.0, "tag": "gemma3:27b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 30.0, "tag": "gemma3:27b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 55.0, "tag": "gemma3:27b-it-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 17.0, "tag": "gemma3:27b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.83, "repo": "mlx-community/gemma-3-27b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-3-27b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 17.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 21.5, "min_memory_gb_q8": 36.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 3 27B Instruct", "value": 4.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Gemma 3 27B Instruct", "value": 42.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 3 27B Instruct", "value": 4.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 3 27B Instruct", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 3 27B Instruct", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 3 27B Instruct", "value": 10.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Gemma 3 27B Instruct", "value": 23.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Gemma 3 27B Instruct", "value": 31.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 3 27B Instruct", "value": 7.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Gemma 3 27B Instruct", "value": 48.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-27b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "gemma-3-27b-it", "value": 1365, "unit": "elo", "ci": 4, "votes": 46666, "rank": 200, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 4.9, "lmarena_elo": 1365 }, "popularity": { "ollama_pulls": "41.2M", "hf_downloads_30d": 413253, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-3-27b-it", "ollama": "https://ollama.com/library/gemma3", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-3-27b-it-4bit" }, "run": { "ollama": "ollama run gemma3:27b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-3-27b-it-4bit --prompt 'hello'" }, "description": "The current, most capable model that runs on a single GPU.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 52.7, "aa_pct_of_frontier": 8.5, "lmarena_elo_behind_frontier": 159, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-3-4b", "name": "Gemma 3 4B", "family": "Gemma 3", "vendor": "Google", "categories": [ "edge", "vision" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-02-20", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 4.3, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "gemma", "name": null, "url": "https://huggingface.co/google/gemma-3-4b-it", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gemma3" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 3.4, "tag": "gemma3:4b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 5.0, "tag": "gemma3:4b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 8.6, "tag": "gemma3:4b-it-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 3.4, "tag": "gemma3:4b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 3.4, "repo": "mlx-community/gemma-3-4b-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-3-4b-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 3.4, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 5.9, "min_memory_gb_q8": 7.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 3 4B Instruct", "value": 4.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Gemma 3 4B Instruct", "value": 29.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 3 4B Instruct", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 3 4B Instruct", "value": 0.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 3 4B Instruct", "value": 0.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 3 4B Instruct", "value": 5.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Gemma 3 4B Instruct", "value": 28.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 3 4B Instruct", "value": 6.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Gemma 3 4B Instruct", "value": 29.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "gemma-3-4b-it", "value": 1303, "unit": "elo", "ci": 9, "votes": 4171, "rank": 284, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 4.8, "lmarena_elo": 1303 }, "popularity": { "ollama_pulls": "41.2M", "hf_downloads_30d": 1138195, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-3-4b-it", "ollama": "https://ollama.com/library/gemma3", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-3-4b-it-4bit" }, "run": { "ollama": "ollama run gemma3:4b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-3-4b-it-4bit --prompt 'hello'" }, "description": "The current, most capable model that runs on a single GPU.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 52.8, "aa_pct_of_frontier": 8.3, "lmarena_elo_behind_frontier": 221, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gemma-3n-e4b", "name": "Gemma 3n E4B", "family": "Gemma 3n", "vendor": "Google", "categories": [ "edge", "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-06-03", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 7.85, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "gemma", "name": null, "url": "https://huggingface.co/google/gemma-3n-E4B-it", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 32768, "ollama_source": "https://ollama.com/library/gemma3n" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 9.8, "tag": "gemma3n:e4b-it-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3n/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.5, "tag": "gemma3n:e4b-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3n/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "gemma3n:e4b-it-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3n/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 9.8, "tag": "gemma3n:e4b", "source_name": "Ollama library", "url": "https://ollama.com/library/gemma3n/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 5.82, "repo": "mlx-community/gemma-3n-E4B-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gemma-3n-E4B-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 9.8, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 13.3, "min_memory_gb_q8": 12.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Gemma 3n E4B Instruct", "value": 4.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Gemma 3n E4B Instruct", "value": 29.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Gemma 3n E4B Instruct", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Gemma 3n E4B Instruct", "value": 2.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Gemma 3n E4B Instruct", "value": 0.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Gemma 3n E4B Instruct", "value": 5.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Gemma 3n E4B Instruct", "value": 27.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Gemma 3n E4B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Gemma 3n E4B Instruct", "value": 26.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gemma-3n-e4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "gemma-3n-e4b-it", "value": 1318, "unit": "elo", "ci": 5, "votes": 22137, "rank": 266, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 4.8, "lmarena_elo": 1318 }, "popularity": { "ollama_pulls": "2.3M", "hf_downloads_30d": 12943, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/gemma-3n-E4B-it", "ollama": "https://ollama.com/library/gemma3n", "mlx_4bit": "https://huggingface.co/mlx-community/gemma-3n-E4B-it-4bit" }, "run": { "ollama": "ollama run gemma3n:e4b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gemma-3n-E4B-it-4bit --prompt 'hello'" }, "description": "Gemma 3n models are designed for efficient execution on everyday devices such as laptops, tablets or phones. ", "notes": [ "Effective-parameter model; Ollama lists q4_K_M (9.8GB) larger than q8_0 (9.5GB) – sizes copied as listed.", "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 52.8, "aa_pct_of_frontier": 8.3, "lmarena_elo_behind_frontier": 206, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "functiongemma-270m", "name": "FunctionGemma 270M", "family": "Gemma 3", "vendor": "Google", "categories": [ "edge", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2025-10-08", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.27, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "gemma", "name": null, "url": "https://huggingface.co/google/functiongemma-270m-it", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 32768, "ollama_source": "https://ollama.com/library/functiongemma" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q8_0": { "value_gb": 0.301, "tag": "functiongemma:270m-it-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/functiongemma/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 0.552, "tag": "functiongemma:270m-it-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/functiongemma/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.301, "tag": "functiongemma:270m", "source_name": "Ollama library", "url": "https://ollama.com/library/functiongemma/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 0.15, "repo": "mlx-community/functiongemma-270m-it-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/functiongemma-270m-it-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.15, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 2.2, "min_memory_gb_q8": 2.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "200.9K", "hf_downloads_30d": 18403, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/functiongemma-270m-it", "ollama": "https://ollama.com/library/functiongemma", "mlx_4bit": "https://huggingface.co/mlx-community/functiongemma-270m-it-4bit" }, "run": { "ollama": "ollama run functiongemma:270m", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/functiongemma-270m-it-4bit --prompt 'hello'" }, "description": "FunctionGemma is a specialized version of Google's Gemma 3 270M model fine-tuned explicitly for function calling. ", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "embeddinggemma-300m", "name": "EmbeddingGemma 300M", "family": "EmbeddingGemma", "vendor": "Google", "categories": [ "embedding" ], "status": "current", "open_weights": true, "release_date": "2025-07-17", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.3, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "gemma", "name": null, "url": "https://huggingface.co/google/embeddinggemma-300m", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 2048, "ollama_source": "https://ollama.com/library/embeddinggemma" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "q8_0": { "value_gb": 0.338, "tag": "embeddinggemma:300m-qat-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 0.622, "tag": "embeddinggemma:300m-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.622, "tag": "embeddinggemma:300m", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 0.17, "repo": "mlx-community/embeddinggemma-300m-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/embeddinggemma-300m-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.17, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 2.2, "min_memory_gb_q8": 2.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "2.5M", "hf_downloads_30d": 3211047, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/google/embeddinggemma-300m", "ollama": "https://ollama.com/library/embeddinggemma", "mlx_4bit": "https://huggingface.co/mlx-community/embeddinggemma-300m-4bit" }, "run": { "ollama": "ollama run embeddinggemma:300m", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/embeddinggemma-300m-4bit --prompt 'hello'" }, "description": "EmbeddingGemma is a 300M parameter embedding model from Google.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "embeddinggemma-2", "name": "EmbeddingGemma 2 (270M-740M)", "family": "EmbeddingGemma", "vendor": "Google", "categories": [ "embedding", "vision", "audio" ], "status": "current", "open_weights": true, "release_date": "2026-09-14", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.74, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/google/embeddinggemma-2", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/embeddinggemma-2" }, "inputs": "Text, Image", "capabilities": [ "vision", "embedding", "audio" ], "files": { "q8_0": { "value_gb": 1.3, "tag": "embeddinggemma-2:740m-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma-2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 1.5, "tag": "embeddinggemma-2:740m-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma-2/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 1.3, "tag": "embeddinggemma-2:740m-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma-2/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 1.4, "tag": "embeddinggemma-2:740m-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma-2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 1.3, "tag": "embeddinggemma-2:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/embeddinggemma-2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 1.3, "weights_basis": "ollama_mlx", "weights_estimated": false, "min_memory_gb": 3.5, "min_memory_gb_q8": 3.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "MTEB", "variant": "*Text* Massive Text Embedding Benchmark (MTEB, multilingual, v2)", "value": 61.36, "unit": "score", "column": "EmbeddingGemma 2", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/google/embeddinggemma-2", "as_of": "2026-10-06", "retrieved": "2026-10-15" }, { "benchmark": "MTEB Code", "variant": "MTEB Code", "value": 78.68, "unit": "score", "column": "EmbeddingGemma 2", "source_type": "vendor-reported", "source_name": "Google Keyword blog (Google DeepMind)", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/embeddinggemma-2/", "as_of": "2026-10-06", "retrieved": "2026-10-16" } ], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "51.2K", "hf_downloads_30d": 151258, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/google/embeddinggemma-2", "ollama": "https://ollama.com/library/embeddinggemma-2", "mlx_4bit": null }, "run": { "ollama": "ollama run embeddinggemma-2", "ollama_cloud_only": false, "mlx": null }, "description": "EmbeddingGemma 2 is a multimodal embedding model from Google built on the Gemma 4 architecture.", "notes": [ "Google's blog says the context window is 8K tokens (about 5.5 minutes of audio, 29 images or 58 video frames). The Hugging Face config and Ollama list 256K.", "Text-only use needs only the 270M part. Ollama's 270m tag is a 304MB to 378MB download (2026-10-16).", "Each result can be cut from 768 numbers to 512, 256 or 128 to save storage, Google says." ], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "muse-glimmer-30b", "name": "Muse Glimmer 30B", "family": "Muse", "vendor": "Meta", "categories": [ "general", "vision", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-08-09", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 29.78, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/muse-glimmer" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 18.0, "tag": "muse-glimmer:30b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 31.0, "tag": "muse-glimmer:30b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 57.0, "tag": "muse-glimmer:30b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 19.0, "tag": "muse-glimmer:30b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 33.0, "tag": "muse-glimmer:30b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 18.0, "tag": "muse-glimmer:30b", "source_name": "Ollama library", "url": "https://ollama.com/library/muse-glimmer/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 19.41, "repo": "mlx-community/Muse-Glimmer-30B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Muse-Glimmer-30B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 18.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 22.7, "min_memory_gb_q8": 37.6, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Muse Glimmer (High)", "value": 17.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Muse Glimmer (High)", "value": 83.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Muse Glimmer (High)", "value": 22.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Muse Glimmer (High)", "value": 51.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Muse Glimmer (High)", "value": 44.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Muse Glimmer (High)", "value": 83.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Muse Glimmer (High)", "value": 74.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-glimmer", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SWE-bench Verified", "variant": "SWE-Bench Verified", "value": 76.0, "unit": "%", "column": "Muse Glimmer-30B\u000bHigh Reasoning", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "as_of": "2026-08-11", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "TerminalBench 2.1 (with terminus2)", "value": 51.7, "unit": "%", "column": "Muse Glimmer-30B\u000bHigh Reasoning", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "as_of": "2026-08-11", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond (AA)", "value": 83.5, "unit": "%", "column": "Muse Glimmer-30B\u000bHigh Reasoning", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "as_of": "2026-08-11", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "muse-glimmer", "value": 1425, "unit": "elo", "ci": 10, "votes": 3897, "rank": 124, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 17.5, "lmarena_elo": 1425 }, "popularity": { "ollama_pulls": "295.3K", "hf_downloads_30d": 252863, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "ollama": "https://ollama.com/library/muse-glimmer", "mlx_4bit": "https://huggingface.co/mlx-community/Muse-Glimmer-30B-4bit" }, "run": { "ollama": "ollama run muse-glimmer:30b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Muse-Glimmer-30B-4bit --prompt 'hello'" }, "description": "Meta's latest open model built for always-on local agents. 30B parameters, licensed under Apache 2.0 and runs on a single GPU — tuned for tool use, long tasks, and failure recovery.", "notes": [ "Meta, Apache-2.0 (per Ollama/HF)." ], "gap": { "aa_points_behind_frontier": 40.1, "aa_pct_of_frontier": 30.4, "lmarena_elo_behind_frontier": 99, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "muse-spark-open-816b-a42b", "name": "Muse Spark Open 816B-A42B", "family": "Muse", "vendor": "Meta", "categories": [ "general", "frontier-open", "vision", "audio", "coding", "agentic", "long-context" ], "status": "current", "open_weights": true, "release_date": "2026-10-15", "release_date_basis": "Meta AI Research blog (weights release); Hugging Face repo was created 2026-09-15", "params": { "total_b": 822.04, "active_b": 42, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"816B total/42B active\" (https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B)" }, "license": { "id": "other", "name": "muse-model-license", "url": "LICENSE.md", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1000000, "native_source": "\"a 1M-token context window\" (https://research.meta.ai/blog/muse-spark-open)", "ollama_default": null, "ollama_source": null }, "inputs": "Text, Image, Audio", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q8_0": { "value_gb": 871.65, "repo": "ggml-org/Muse-Spark-Open-816B-A42B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/ggml-org/Muse-Spark-Open-816B-A42B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 461.7, "repo": "mlx-community/Muse-Spark-Open-816B-A42B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Muse-Spark-Open-816B-A42B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 461.7, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 533.0, "min_memory_gb_q8": 1004.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Muse Spark Open (Xhigh)", "value": 38.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Muse Spark Open (Xhigh)", "value": 89.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Muse Spark Open (Xhigh)", "value": 43.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Muse Spark Open (Xhigh)", "value": 76.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Muse Spark Open (Xhigh)", "value": 58.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Muse Spark Open (Xhigh)", "value": 78.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Muse Spark Open (Xhigh)", "value": 80.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/muse-spark-open", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 38.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.1K", "hf_downloads_30d": 815, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "ollama": "https://ollama.com/library/muse-spark-open", "mlx_4bit": "https://huggingface.co/mlx-community/Muse-Spark-Open-816B-A42B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/Muse-Spark-Open-816B-A42B-4bit --prompt 'hello'" }, "description": "Meta's biggest open model: 816B in total, 42B active per word, built for coding and agents, reading text, pictures and sound.", "notes": [ "Meta says it runs on two M5 Ultra Mac Studios with 512 GB each, using MLX. Far too big for one 128 GB Mac.", "On Ollama it is cloud only (muse-spark-open:cloud).", "Muse Model License with limited commercial terms; Hugging Face asks for approval before download.", "Audio and audio-video input per the PyTorch blog (https://pytorch.org/blog/muse-spark-open/)." ], "gap": { "aa_points_behind_frontier": 18.7, "aa_pct_of_frontier": 67.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "llama-4-scout", "name": "Llama 4 Scout 17B-16E (109B)", "family": "Llama 4", "vendor": "Meta", "categories": [ "general", "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-04-02", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 108.64, "active_b": 17, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"17B (Activated)\", 16 experts (https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct)" }, "license": { "id": "other", "name": "llama4", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 10485760, "ollama_source": "https://ollama.com/library/llama4" }, "inputs": null, "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 67.0, "tag": "llama4:17b-scout-16e-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 117.0, "tag": "llama4:17b-scout-16e-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 217.0, "tag": "llama4:17b-scout-16e-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 61.12, "repo": "mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 67.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 79.0, "min_memory_gb_q8": 136.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Llama 4 Scout", "value": 6.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Llama 4 Scout", "value": 58.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Llama 4 Scout", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Llama 4 Scout", "value": 1.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Llama 4 Scout", "value": 3.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Llama 4 Scout", "value": 15.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Llama 4 Scout", "value": 21.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Llama 4 Scout", "value": 39.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Llama 4 Scout", "value": 27.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Llama 4 Scout", "value": 52.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-scout", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "llama-4-scout-17b-16e-instruct", "value": 1322, "unit": "elo", "ci": 5, "votes": 29826, "rank": 260, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 6.5, "lmarena_elo": 1322 }, "popularity": { "ollama_pulls": "1.9M", "hf_downloads_30d": 164332, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "ollama": "https://ollama.com/library/llama4", "mlx_4bit": "https://huggingface.co/mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit" }, "run": { "ollama": "ollama run llama4:17b-scout", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit --prompt 'hello'" }, "description": "Meta's latest collection of multimodal models.", "notes": [ "Ollama lists a 10M-token context for Scout." ], "gap": { "aa_points_behind_frontier": 51.1, "aa_pct_of_frontier": 11.3, "lmarena_elo_behind_frontier": 202, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "llama-4-maverick", "name": "Llama 4 Maverick 17B-128E (400B)", "family": "Llama 4", "vendor": "Meta", "categories": [ "general", "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-04-01", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 401.58, "active_b": 17, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"17B (Activated)\", 128 experts (https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct)" }, "license": { "id": "other", "name": "llama4", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 10485760, "ollama_source": "https://ollama.com/library/llama4" }, "inputs": null, "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 245.0, "tag": "llama4:17b-maverick-128e-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 428.0, "tag": "llama4:17b-maverick-128e-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 803.0, "tag": "llama4:17b-maverick-128e-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/llama4/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 225.9, "repo": "mlx-community/Llama-4-Maverick-17B-128E-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Llama-4-Maverick-17B-128E-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 245.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 283.8, "min_memory_gb_q8": 494.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Llama 4 Maverick", "value": 9.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Llama 4 Maverick", "value": 67.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Llama 4 Maverick", "value": 4.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Llama 4 Maverick", "value": 6.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Llama 4 Maverick", "value": 7.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Llama 4 Maverick", "value": 17.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Llama 4 Maverick", "value": 31.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Llama 4 Maverick", "value": 43.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Llama 4 Maverick", "value": 50.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Llama 4 Maverick", "value": 62.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-4-maverick", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "llama-4-maverick-17b-128e-instruct", "value": 1327, "unit": "elo", "ci": 4, "votes": 39450, "rank": 249, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 9.3, "lmarena_elo": 1327 }, "popularity": { "ollama_pulls": "1.9M", "hf_downloads_30d": 9442, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "ollama": "https://ollama.com/library/llama4", "mlx_4bit": "https://huggingface.co/mlx-community/Llama-4-Maverick-17B-128E-Instruct-4bit" }, "run": { "ollama": "ollama run llama4:17b-maverick", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Llama-4-Maverick-17B-128E-Instruct-4bit --prompt 'hello'" }, "description": "Meta's latest collection of multimodal models.", "notes": [], "gap": { "aa_points_behind_frontier": 48.3, "aa_pct_of_frontier": 16.1, "lmarena_elo_behind_frontier": 197, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "llama-3.3-70b", "name": "Llama 3.3 70B", "family": "Llama 3", "vendor": "Meta", "categories": [ "general" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-11-26", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 70.55, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "llama3.3", "name": null, "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 131072, "ollama_source": "https://ollama.com/library/llama3.3" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 43.0, "tag": "llama3.3:70b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 75.0, "tag": "llama3.3:70b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 141.0, "tag": "llama3.3:70b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 43.0, "tag": "llama3.3:70b", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 39.69, "repo": "mlx-community/Llama-3.3-70B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Llama-3.3-70B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 43.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 51.4, "min_memory_gb_q8": 88.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Llama 3.3 Instruct 70B", "value": 7.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Llama 3.3 Instruct 70B", "value": 49.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Llama 3.3 Instruct 70B", "value": 3.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Llama 3.3 Instruct 70B", "value": 3.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Llama 3.3 Instruct 70B", "value": 4.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Llama 3.3 Instruct 70B", "value": 26.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Llama 3.3 Instruct 70B", "value": 47.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Llama 3.3 Instruct 70B", "value": 15.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-3-instruct-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Reasoning GPQA Diamond (CoT)", "value": 50.5, "unit": "%", "column": "Llama-3.3 70B Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "as_of": "2024-12-21", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "llama-3.3-70b-instruct", "value": 1318, "unit": "elo", "ci": 4, "votes": 54497, "rank": 264, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 7.7, "lmarena_elo": 1318 }, "popularity": { "ollama_pulls": "4.2M", "hf_downloads_30d": 265985, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "ollama": "https://ollama.com/library/llama3.3", "mlx_4bit": "https://huggingface.co/mlx-community/Llama-3.3-70B-Instruct-4bit" }, "run": { "ollama": "ollama run llama3.3:70b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Llama-3.3-70B-Instruct-4bit --prompt 'hello'" }, "description": "New state of the art 70B model. Llama 3.3 70B offers similar performance compared to the Llama 3.1 405B model.", "notes": [], "gap": { "aa_points_behind_frontier": 49.9, "aa_pct_of_frontier": 13.4, "lmarena_elo_behind_frontier": 206, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "llama-3.1-8b", "name": "Llama 3.1 8B", "family": "Llama 3", "vendor": "Meta", "categories": [ "general" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-07-18", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.03, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "llama3.1", "name": null, "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 131072, "ollama_source": "https://ollama.com/library/llama3.1" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 4.9, "tag": "llama3.1:8b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 8.5, "tag": "llama3.1:8b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "llama3.1:8b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 4.9, "tag": "llama3.1:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 4.52, "repo": "mlx-community/Llama-3.1-8B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Llama-3.1-8B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 4.9, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.6, "min_memory_gb_q8": 11.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Llama 3.1 Instruct 8B", "value": 6.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Llama 3.1 Instruct 8B", "value": 25.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Llama 3.1 Instruct 8B", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Llama 3.1 Instruct 8B", "value": 0.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Llama 3.1 Instruct 8B", "value": 1.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Llama 3.1 Instruct 8B", "value": 16.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Llama 3.1 Instruct 8B", "value": 28.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Llama 3.1 Instruct 8B", "value": 18.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-1-instruct-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro (CoT)", "value": 55.0, "unit": "%", "column": "Llama 3.1 8B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "as_of": "2024-09-25", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 39.5, "unit": "%", "column": "Llama 3.1 8B Instruct", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "as_of": "2024-09-25", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "llama-3.1-8b-instruct", "value": 1211, "unit": "elo", "ci": 4, "votes": 49605, "rank": 343, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 6.9, "lmarena_elo": 1211 }, "popularity": { "ollama_pulls": "120.3M", "hf_downloads_30d": 5930556, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "ollama": "https://ollama.com/library/llama3.1", "mlx_4bit": "https://huggingface.co/mlx-community/Llama-3.1-8B-Instruct-4bit" }, "run": { "ollama": "ollama run llama3.1:8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Llama-3.1-8B-Instruct-4bit --prompt 'hello'" }, "description": "Llama 3.1 is a new state-of-the-art model from Meta available in 8B, 70B and 405B parameter sizes.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 50.7, "aa_pct_of_frontier": 12.0, "lmarena_elo_behind_frontier": 313, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "llama-3.2-3b", "name": "Llama 3.2 3B", "family": "Llama 3", "vendor": "Meta", "categories": [ "edge" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-09-18", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.21, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "llama3.2", "name": null, "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "gated": "manual", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 131072, "ollama_source": "https://ollama.com/library/llama3.2" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 2.0, "tag": "llama3.2:3b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 3.4, "tag": "llama3.2:3b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 6.4, "tag": "llama3.2:3b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 2.0, "tag": "llama3.2:3b", "source_name": "Ollama library", "url": "https://ollama.com/library/llama3.2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 1.81, "repo": "mlx-community/Llama-3.2-3B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Llama-3.2-3B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 2.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 4.3, "min_memory_gb_q8": 5.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Llama 3.2 Instruct 3B", "value": 5.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Llama 3.2 Instruct 3B", "value": 25.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Llama 3.2 Instruct 3B", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Llama 3.2 Instruct 3B", "value": 21.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Llama 3.2 Instruct 3B", "value": 26.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Llama 3.2 Instruct 3B", "value": 4.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/llama-3-2-instruct-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "llama-3.2-3b-instruct", "value": 1167, "unit": "elo", "ci": 8, "votes": 7936, "rank": 369, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 5.7, "lmarena_elo": 1167 }, "popularity": { "ollama_pulls": "85.5M", "hf_downloads_30d": 1308673, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "ollama": "https://ollama.com/library/llama3.2", "mlx_4bit": "https://huggingface.co/mlx-community/Llama-3.2-3B-Instruct-4bit" }, "run": { "ollama": "ollama run llama3.2:3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Llama-3.2-3B-Instruct-4bit --prompt 'hello'" }, "description": "Meta's Llama 3.2 goes small with 1B and 3B models. ", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 51.9, "aa_pct_of_frontier": 9.9, "lmarena_elo_behind_frontier": 357, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mistral-large-4", "name": "Mistral Large 4", "family": "Mistral", "vendor": "Mistral AI", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": null, "release_date_basis": null, "params": { "total_b": 1050, "active_b": 52, "is_moe": true, "total_source": "Ollama description: \"52B of 1.05T total parameters\"", "active_source": "\"activates 52B of 1.05T total parameters\" (https://ollama.com/library/mistral-large-4)" }, "license": { "id": null, "name": null, "url": null, "gated": false, "source": null }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/mistral-large-4" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": {}, "memory": { "weights_gb_4bit": 630.0, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 726.5, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "LMArena Text Elo", "variant": "mistral-large-4", "value": 1432, "unit": "elo", "ci": 7, "votes": 7451, "rank": 111, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": null, "lmarena_elo": 1432 }, "popularity": { "ollama_pulls": "7,736", "hf_downloads_30d": null, "retrieved": "2026-10-15" }, "links": { "huggingface": null, "ollama": "https://ollama.com/library/mistral-large-4", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "Mistral Large 4 is Mistral's open-weight, general-purpose multimodal model. Its granular Mixture-of-Experts design activates 52B of 1.05T total parameters, with a 1.6B vision encoder handling images.", "notes": [ "Ollama lists it as open-weight but cloud-only; LMArena lists the license as Proprietary and no Hugging Face repo was found under mistralai on 2026-10-15. Verify before listing as open-weight.", "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows.", "Mistral's launch page (2026-10-06, opened 2026-10-16) calls Large 4 a public preview: 1 trillion parameters, 52B active, paid preview API now, weights promised by the end of October 2026. No Hugging Face repo under mistralai on 2026-10-16." ], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": 92, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mistral-large-3", "name": "Mistral Large 3 675B", "family": "Mistral", "vendor": "Mistral AI", "categories": [ "general", "vision" ], "status": "superseded", "open_weights": true, "release_date": "2025-11-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 675.0, "active_b": 41, "is_moe": true, "total_source": "parsed from model name", "active_source": "\"41B active parameters and 675B total parameters\" (https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 262144, "ollama_source": "https://ollama.com/library/mistral-large-3" }, "inputs": null, "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 406.99, "repo": "unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 715.68, "repo": "unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 386.73, "repo": "mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 406.99, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 470.0, "min_memory_gb_q8": 825.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Mistral Large 3", "value": 9.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Mistral Large 3", "value": 68.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Mistral Large 3", "value": 4.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Mistral Large 3", "value": 15.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Mistral Large 3", "value": 12.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Mistral Large 3", "value": 24.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Mistral Large 3", "value": 36.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Mistral Large 3", "value": 36.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Mistral Large 3", "value": 36.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Mistral Large 3", "value": 55.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-large-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "mistral-large-3", "value": 1414, "unit": "elo", "ci": 3, "votes": 84205, "rank": 145, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 9.3, "lmarena_elo": 1414 }, "popularity": { "ollama_pulls": "136.6K", "hf_downloads_30d": 2211, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512", "ollama": "https://ollama.com/library/mistral-large-3", "mlx_4bit": "https://huggingface.co/mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit --prompt 'hello'" }, "description": "A general-purpose multimodal mixture-of-experts model for production-grade tasks and enterprise workloads.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows.", "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 48.3, "aa_pct_of_frontier": 16.1, "lmarena_elo_behind_frontier": 110, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mistral-medium-3.5-128b", "name": "Mistral Medium 3.5 128B", "family": "Mistral", "vendor": "Mistral AI", "categories": [ "general", "vision", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2026-03-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 127.7, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": null, "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/mistral-medium-3.5" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 80.0, "tag": "mistral-medium-3.5:128b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-medium-3.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 138.0, "tag": "mistral-medium-3.5:128b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-medium-3.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 255.0, "tag": "mistral-medium-3.5:128b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-medium-3.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 80.0, "tag": "mistral-medium-3.5:128b", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-medium-3.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 78.0, "repo": "mlx-community/Mistral-Medium-3.5-128B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Mistral-Medium-3.5-128B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 80.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 94.0, "min_memory_gb_q8": 160.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Mistral Medium 3.5", "value": 14.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Mistral Medium 3.5", "value": 74.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Mistral Medium 3.5", "value": 13.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Mistral Medium 3.5", "value": 33.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Mistral Medium 3.5", "value": 50.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Mistral Medium 3.5", "value": 94.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Mistral Medium 3.5", "value": 40.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Mistral Medium 3.5", "value": 68.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Mistral Medium 3.5", "value": 69.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Mistral Medium 3.5", "value": 64.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-medium-3-5", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "mistral-medium-3.5", "value": 1426, "unit": "elo", "ci": 6, "votes": 11778, "rank": 117, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 14.2, "lmarena_elo": 1426 }, "popularity": { "ollama_pulls": "427.5K", "hf_downloads_30d": 73667, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "ollama": "https://ollama.com/library/mistral-medium-3.5", "mlx_4bit": "https://huggingface.co/mlx-community/Mistral-Medium-3.5-128B-4bit" }, "run": { "ollama": "ollama run mistral-medium-3.5:128b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Mistral-Medium-3.5-128B-4bit --prompt 'hello'" }, "description": "Mistral Medium 3.5 is the first flagship model of Mistral AI that merged instruction-following, reasoning, and coding in a single set of 128B weights.", "notes": [], "gap": { "aa_points_behind_frontier": 43.4, "aa_pct_of_frontier": 24.7, "lmarena_elo_behind_frontier": 98, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mistral-small-4-119b", "name": "Mistral Small 4 119B", "family": "Mistral", "vendor": "Mistral AI", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-01-23", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 119.4, "active_b": 6.5, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"119B parameters, with 6.5B activated per token\" (https://huggingface.co/mistralai/Mistral-Small-4-119B-2603)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 72.16, "repo": "lmstudio-community/Mistral-Small-4-119B-2603-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/lmstudio-community/Mistral-Small-4-119B-2603-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 126.47, "repo": "unsloth/Mistral-Small-4-119B-2603-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Mistral-Small-4-119B-2603-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 67.75, "repo": "mlx-community/Mistral-Small-4-119B-2603-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Mistral-Small-4-119B-2603-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 72.16, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 85.0, "min_memory_gb_q8": 147.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Mistral Small 4 (Reasoning)", "value": 11.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Mistral Small 4 (Reasoning)", "value": 76.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Mistral Small 4 (Reasoning)", "value": 9.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Mistral Small 4 (Reasoning)", "value": 17.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Mistral Small 4 (Reasoning)", "value": 21.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Mistral Small 4 (Reasoning)", "value": 41.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Mistral Small 4 (Reasoning)", "value": 38.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Mistral Small 4 (Reasoning)", "value": 48.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Mistral Small 4 (Reasoning)", "value": 49.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Mistral Small 4 (Reasoning)", "value": 56.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Mistral Small 4 (Non-reasoning)", "value": 9.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Mistral Small 4 (Non-reasoning)", "value": 57.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Mistral Small 4 (Non-reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Mistral Small 4 (Non-reasoning)", "value": 10.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Mistral Small 4 (Non-reasoning)", "value": 18.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Mistral Small 4 (Non-reasoning)", "value": 32.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Mistral Small 4 (Non-reasoning)", "value": 28.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Mistral Small 4 (Non-reasoning)", "value": 46.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-4-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 11.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 74916, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/Mistral-Small-4-119B-2603-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/Mistral-Small-4-119B-2603-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 46.3, "aa_pct_of_frontier": 19.6, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mistral-small-3.2-24b", "name": "Mistral Small 3.2 24B", "family": "Mistral", "vendor": "Mistral AI", "categories": [ "general", "vision" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-06-19", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 24.01, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/mistral-small3.2" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 15.0, "tag": "mistral-small3.2:24b-instruct-2506-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-small3.2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 26.0, "tag": "mistral-small3.2:24b-instruct-2506-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-small3.2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 48.0, "tag": "mistral-small3.2:24b-instruct-2506-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-small3.2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 15.0, "tag": "mistral-small3.2:24b", "source_name": "Ollama library", "url": "https://ollama.com/library/mistral-small3.2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 13.26, "repo": "mlx-community/Mistral-Small-3.2-24B-Instruct-2506-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Mistral-Small-3.2-24B-Instruct-2506-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 15.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 19.2, "min_memory_gb_q8": 31.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Mistral Small 3.2", "value": 7.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Mistral Small 3.2", "value": 50.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Mistral Small 3.2", "value": 4.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Mistral Small 3.2", "value": 6.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Mistral Small 3.2", "value": 5.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Mistral Small 3.2", "value": 29.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Mistral Small 3.2", "value": 28.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Mistral Small 3.2", "value": 33.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Mistral Small 3.2", "value": 20.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Mistral Small 3.2", "value": 48.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mistral-small-3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "mistral-small-2506", "value": 1356, "unit": "elo", "ci": 5, "votes": 17462, "rank": 210, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 7.0, "lmarena_elo": 1356 }, "popularity": { "ollama_pulls": "2.5M", "hf_downloads_30d": 248354, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "ollama": "https://ollama.com/library/mistral-small3.2", "mlx_4bit": "https://huggingface.co/mlx-community/Mistral-Small-3.2-24B-Instruct-2506-4bit" }, "run": { "ollama": "ollama run mistral-small3.2:24b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Mistral-Small-3.2-24B-Instruct-2506-4bit --prompt 'hello'" }, "description": "An update to Mistral Small that improves on function calling, instruction following, and less repetition errors.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 50.6, "aa_pct_of_frontier": 12.2, "lmarena_elo_behind_frontier": 168, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "magistral-small-2509", "name": "Magistral Small 1.2 24B", "family": "Magistral", "vendor": "Mistral AI", "categories": [ "reasoning", "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-09-12", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 24.01, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Magistral-Small-2509", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 39936, "ollama_source": "https://ollama.com/library/magistral" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 14.0, "tag": "magistral:24b-small-2506-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/magistral/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 25.0, "tag": "magistral:24b-small-2506-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/magistral/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 47.0, "tag": "magistral:24b-small-2506-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/magistral/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 14.0, "tag": "magistral:24b", "source_name": "Ollama library", "url": "https://ollama.com/library/magistral/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 14.09, "repo": "lmstudio-community/Magistral-Small-2509-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/lmstudio-community/Magistral-Small-2509-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 14.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 18.1, "min_memory_gb_q8": 30.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Magistral Small 1.2", "value": 8.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Magistral Small 1.2", "value": 66.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Magistral Small 1.2", "value": 6.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Magistral Small 1.2", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Magistral Small 1.2", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Magistral Small 1.2", "value": 27.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Magistral Small 1.2", "value": 44.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Magistral Small 1.2", "value": 19.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Magistral Small 1.2", "value": 55.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/magistral-small-2509", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 8.6, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.5M", "hf_downloads_30d": 6859, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Magistral-Small-2509", "ollama": "https://ollama.com/library/magistral", "mlx_4bit": "https://huggingface.co/lmstudio-community/Magistral-Small-2509-MLX-4bit" }, "run": { "ollama": "ollama run magistral:24b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model lmstudio-community/Magistral-Small-2509-MLX-4bit --prompt 'hello'" }, "description": "Magistral is a small, efficient reasoning model with 24B parameters.", "notes": [], "gap": { "aa_points_behind_frontier": 49.0, "aa_pct_of_frontier": 14.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "devstral-small-2-24b", "name": "Devstral Small 2 24B", "family": "Devstral", "vendor": "Mistral AI", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2025-11-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 24.01, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 393216, "native_source": "HF config.json", "ollama_default": 393216, "ollama_source": "https://ollama.com/library/devstral-small-2" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 15.0, "tag": "devstral-small-2:24b-instruct-2512-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-small-2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 26.0, "tag": "devstral-small-2:24b-instruct-2512-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-small-2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 48.0, "tag": "devstral-small-2:24b-instruct-2512-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-small-2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 15.0, "tag": "devstral-small-2:24b", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-small-2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 15.1, "repo": "mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 15.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 19.2, "min_memory_gb_q8": 31.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Devstral Small 2", "value": 7.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Devstral Small 2", "value": 53.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Devstral Small 2", "value": 3.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Devstral Small 2", "value": 16.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Devstral Small 2", "value": 29.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Devstral Small 2", "value": 23.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Devstral Small 2", "value": 32.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Devstral Small 2", "value": 31.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Devstral Small 2", "value": 28.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Devstral Small 2", "value": 44.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-small-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 7.5, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.8M", "hf_downloads_30d": 312196, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512", "ollama": "https://ollama.com/library/devstral-small-2", "mlx_4bit": "https://huggingface.co/mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit" }, "run": { "ollama": "ollama run devstral-small-2:24b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Devstral-Small-2-24B-Instruct-2512-4bit --prompt 'hello'" }, "description": "24B model that excels at using tools to explore codebases, editing multiple files and power software engineering agents.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 50.1, "aa_pct_of_frontier": 13.0, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "devstral-2-123b", "name": "Devstral 2 123B", "family": "Devstral", "vendor": "Mistral AI", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2025-11-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 125.03, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": null, "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/devstral-2" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 75.0, "tag": "devstral-2:123b-instruct-2512-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 133.0, "tag": "devstral-2:123b-instruct-2512-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 250.0, "tag": "devstral-2:123b-instruct-2512-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 75.0, "tag": "devstral-2:123b", "source_name": "Ollama library", "url": "https://ollama.com/library/devstral-2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 70.33, "repo": "mlx-community/Devstral-2-123B-Instruct-2512-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Devstral-2-123B-Instruct-2512-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 75.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 88.2, "min_memory_gb_q8": 154.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Devstral 2", "value": 8.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Devstral 2", "value": 59.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Devstral 2", "value": 3.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Devstral 2", "value": 18.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Devstral 2", "value": 30.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Devstral 2", "value": 24.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Devstral 2", "value": 32.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Devstral 2", "value": 38.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Devstral 2", "value": 32.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/devstral-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 8.6, "lmarena_elo": null }, "popularity": { "ollama_pulls": "363.3K", "hf_downloads_30d": 5080, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512", "ollama": "https://ollama.com/library/devstral-2", "mlx_4bit": "https://huggingface.co/mlx-community/Devstral-2-123B-Instruct-2512-4bit" }, "run": { "ollama": "ollama run devstral-2:123b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Devstral-2-123B-Instruct-2512-4bit --prompt 'hello'" }, "description": "123B model that excels at using tools to explore codebases, editing multiple files and power software engineering agents.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 49.0, "aa_pct_of_frontier": 14.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "ministral-3-3b", "name": "Ministral 3 3B", "family": "Ministral 3", "vendor": "Mistral AI", "categories": [ "edge", "vision" ], "status": "current", "open_weights": true, "release_date": "2025-10-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.85, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/ministral-3" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 3.0, "tag": "ministral-3:3b-instruct-2512-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 4.5, "tag": "ministral-3:3b-instruct-2512-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 7.7, "tag": "ministral-3:3b-instruct-2512-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 3.0, "tag": "ministral-3:3b", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 2.75, "repo": "mlx-community/Ministral-3-3B-Instruct-2512-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Ministral-3-3B-Instruct-2512-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 3.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 5.4, "min_memory_gb_q8": 7.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Ministral 3 3B", "value": 4.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Ministral 3 3B", "value": 35.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Ministral 3 3B", "value": 5.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Ministral 3 3B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Ministral 3 3B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Ministral 3 3B", "value": 24.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Ministral 3 3B", "value": 15.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Ministral 3 3B", "value": 26.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Ministral 3 3B", "value": 17.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Ministral 3 3B", "value": 38.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 4.8, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.5M", "hf_downloads_30d": 107866, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512", "ollama": "https://ollama.com/library/ministral-3", "mlx_4bit": "https://huggingface.co/mlx-community/Ministral-3-3B-Instruct-2512-4bit" }, "run": { "ollama": "ollama run ministral-3:3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Ministral-3-3B-Instruct-2512-4bit --prompt 'hello'" }, "description": "The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware.", "notes": [], "gap": { "aa_points_behind_frontier": 52.8, "aa_pct_of_frontier": 8.3, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "ministral-3-8b", "name": "Ministral 3 8B", "family": "Ministral 3", "vendor": "Mistral AI", "categories": [ "edge", "vision" ], "status": "current", "open_weights": true, "release_date": "2025-10-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.92, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/ministral-3" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 6.0, "tag": "ministral-3:8b-instruct-2512-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.9, "tag": "ministral-3:8b-instruct-2512-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 18.0, "tag": "ministral-3:8b-instruct-2512-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 6.0, "tag": "ministral-3:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 5.6, "repo": "mlx-community/Ministral-3-8B-Instruct-2512-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Ministral-3-8B-Instruct-2512-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 6.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.9, "min_memory_gb_q8": 13.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Ministral 3 8B", "value": 5.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Ministral 3 8B", "value": 47.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Ministral 3 8B", "value": 4.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Ministral 3 8B", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Ministral 3 8B", "value": 4.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Ministral 3 8B", "value": 26.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Ministral 3 8B", "value": 20.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Ministral 3 8B", "value": 29.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Ministral 3 8B", "value": 25.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Ministral 3 8B", "value": 46.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 5.5, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.5M", "hf_downloads_30d": 116423, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512", "ollama": "https://ollama.com/library/ministral-3", "mlx_4bit": "https://huggingface.co/mlx-community/Ministral-3-8B-Instruct-2512-4bit" }, "run": { "ollama": "ollama run ministral-3:8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Ministral-3-8B-Instruct-2512-4bit --prompt 'hello'" }, "description": "The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware.", "notes": [], "gap": { "aa_points_behind_frontier": 52.1, "aa_pct_of_frontier": 9.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "ministral-3-14b", "name": "Ministral 3 14B", "family": "Ministral 3", "vendor": "Mistral AI", "categories": [ "general", "vision" ], "status": "current", "open_weights": true, "release_date": "2025-10-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 13.95, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/ministral-3" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q4_k_m": { "value_gb": 9.1, "tag": "ministral-3:14b-instruct-2512-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 15.0, "tag": "ministral-3:14b-instruct-2512-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 28.0, "tag": "ministral-3:14b-instruct-2512-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 9.1, "tag": "ministral-3:14b", "source_name": "Ollama library", "url": "https://ollama.com/library/ministral-3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 8.42, "repo": "mlx-community/Ministral-3-14B-Instruct-2512-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Ministral-3-14B-Instruct-2512-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 9.1, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 12.5, "min_memory_gb_q8": 19.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Ministral 3 14B", "value": 6.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Ministral 3 14B", "value": 57.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Ministral 3 14B", "value": 4.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Ministral 3 14B", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Ministral 3 14B", "value": 9.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Ministral 3 14B", "value": 27.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Ministral 3 14B", "value": 23.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Ministral 3 14B", "value": 32.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Ministral 3 14B", "value": 26.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Ministral 3 14B", "value": 49.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/ministral-3-14b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 6.0, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.5M", "hf_downloads_30d": 115986, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "ollama": "https://ollama.com/library/ministral-3", "mlx_4bit": "https://huggingface.co/mlx-community/Ministral-3-14B-Instruct-2512-4bit" }, "run": { "ollama": "ollama run ministral-3:14b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Ministral-3-14B-Instruct-2512-4bit --prompt 'hello'" }, "description": "The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware.", "notes": [], "gap": { "aa_points_behind_frontier": 51.6, "aa_pct_of_frontier": 10.4, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "deepseek-v4.1-flash", "name": "DeepSeek V4.1 Flash", "family": "DeepSeek V4", "vendor": "DeepSeek", "categories": [ "general", "coding", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-09-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 763.21, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/deepseek-v4.1-flash" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": { "mlx_4bit": { "value_gb": 552.86, "repo": "mlx-community/DeepSeek-V4.1-Flash-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/DeepSeek-V4.1-Flash-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 552.86, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 637.8, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V4.1 Flash (Max)", "value": 39.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V4.1 Flash (Max)", "value": 39.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "DeepSeek V4.1 Flash (Max)", "value": 51.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V4.1 Flash (Max)", "value": 84.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "DeepSeek V4.1 Flash (Max)", "value": 77.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V4.1 Flash (Non-reasoning)", "value": 24.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V4.1 Flash (Non-reasoning)", "value": 10.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "DeepSeek V4.1 Flash (Non-reasoning)", "value": 35.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V4.1 Flash (Non-reasoning)", "value": 55.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "DeepSeek V4.1 Flash (Non-reasoning)", "value": 65.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "deepseek-v4.1-flash-max", "value": 1475, "unit": "elo", "ci": 6, "votes": 12793, "rank": 38, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 39.5, "lmarena_elo": 1475 }, "popularity": { "ollama_pulls": "131.3K", "hf_downloads_30d": 1238583, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "ollama": "https://ollama.com/library/deepseek-v4.1-flash", "mlx_4bit": "https://huggingface.co/mlx-community/DeepSeek-V4.1-Flash-MLX-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/DeepSeek-V4.1-Flash-MLX-4bit --prompt 'hello'" }, "description": "DeepSeek-V4.1-Flash is an advanced tool designed to enhance search capabilities, providing users with faster and more accurate results.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 18.1, "aa_pct_of_frontier": 68.6, "lmarena_elo_behind_frontier": 49, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "deepseek-v4-pro", "name": "DeepSeek V4 Pro (0813)", "family": "DeepSeek V4", "vendor": "DeepSeek", "categories": [ "general", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-08-13", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1650.5, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/deepseek-v4-pro" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": {}, "memory": { "weights_gb_4bit": 990.3, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 1140.8, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 36.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 92.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 41.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 78.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 51.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V4 Pro 0813 (Max)", "value": 80.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V4 Pro 0813 (Non-reasoning)", "value": 20.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V4 Pro 0813 (Non-reasoning)", "value": 10.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "DeepSeek V4 Pro 0813 (Non-reasoning)", "value": 39.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V4 Pro 0813 (Non-reasoning)", "value": 50.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "deepseek-v4-pro-high-20260813", "value": 1465, "unit": "elo", "ci": 7, "votes": 9800, "rank": 56, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 36.0, "lmarena_elo": 1465 }, "popularity": { "ollama_pulls": "456.1K", "hf_downloads_30d": 141935, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "ollama": "https://ollama.com/library/deepseek-v4-pro", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "DeepSeek-V4-Pro is a frontier Mixture-of-Experts model with a large context window and three reasoning modes.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 21.6, "aa_pct_of_frontier": 62.5, "lmarena_elo_behind_frontier": 59, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "deepseek-v3.2", "name": "DeepSeek V3.2", "family": "DeepSeek V3", "vendor": "DeepSeek", "categories": [ "general" ], "status": "legacy", "open_weights": true, "release_date": "2025-12-01", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 685.36, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 163840, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 405.4, "repo": "unsloth/DeepSeek-V3.2-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/DeepSeek-V3.2-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 713.29, "repo": "unsloth/DeepSeek-V3.2-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/DeepSeek-V3.2-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 378.09, "repo": "mlx-community/DeepSeek-V3.2-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/DeepSeek-V3.2-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 405.4, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 468.2, "min_memory_gb_q8": 822.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V3.2 (Reasoning)", "value": 21.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "DeepSeek V3.2 (Reasoning)", "value": 84.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V3.2 (Reasoning)", "value": 24.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "DeepSeek V3.2 (Reasoning)", "value": 35.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "DeepSeek V3.2 (Reasoning)", "value": 46.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "DeepSeek V3.2 (Reasoning)", "value": 90.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "DeepSeek V3.2 (Reasoning)", "value": 60.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V3.2 (Reasoning)", "value": 73.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 16.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 75.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 11.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 32.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 78.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 49.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek V3.2 (Non-reasoning)", "value": 45.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-v3-2", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "deepseek-v3.2", "value": 1425, "unit": "elo", "ci": 3, "votes": 48112, "rank": 123, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 21.5, "lmarena_elo": 1425 }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 4927838, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/DeepSeek-V3.2-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/DeepSeek-V3.2-4bit --prompt 'hello'" }, "description": null, "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 36.1, "aa_pct_of_frontier": 37.3, "lmarena_elo_behind_frontier": 99, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "deepseek-r1-0528-qwen3-8b", "name": "DeepSeek-R1-0528-Qwen3-8B", "family": "DeepSeek R1", "vendor": "DeepSeek", "categories": [ "reasoning" ], "status": "legacy-popular", "open_weights": true, "release_date": "2025-05-29", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.19, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/deepseek-r1" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 5.2, "tag": "deepseek-r1:8b-0528-qwen3-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-r1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 8.9, "tag": "deepseek-r1:8b-0528-qwen3-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-r1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "deepseek-r1:8b-0528-qwen3-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-r1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 5.2, "tag": "deepseek-r1:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-r1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 4.61, "repo": "mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.2, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.0, "min_memory_gb_q8": 12.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 8.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 61.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 5.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 1.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 19.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "DeepSeek R1 0528 Qwen3 8B", "value": 15.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true } ], "summary": { "aa_index_best": 8.1, "lmarena_elo": null }, "popularity": { "ollama_pulls": "93.7M", "hf_downloads_30d": 525777, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", "ollama": "https://ollama.com/library/deepseek-r1", "mlx_4bit": "https://huggingface.co/mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit" }, "run": { "ollama": "ollama run deepseek-r1:8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit --prompt 'hello'" }, "description": "DeepSeek-R1 is a family of open reasoning models with performance approaching that of leading models, such as O3 and Gemini 2.5 Pro.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 49.5, "aa_pct_of_frontier": 14.1, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "deepseek-ocr", "name": "DeepSeek-OCR 3B", "family": "DeepSeek OCR", "vendor": "DeepSeek", "categories": [ "vision", "ocr" ], "status": "current", "open_weights": true, "release_date": "2025-10-17", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.34, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 8192, "native_source": "HF config.json", "ollama_default": 8192, "ollama_source": "https://ollama.com/library/deepseek-ocr" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "bf16": { "value_gb": 6.7, "tag": "deepseek-ocr:3b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-ocr/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 6.7, "tag": "deepseek-ocr:3b", "source_name": "Ollama library", "url": "https://ollama.com/library/deepseek-ocr/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 2.45, "repo": "mlx-community/DeepSeek-OCR-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/DeepSeek-OCR-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 2.45, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 4.8, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "553.4K", "hf_downloads_30d": 2040609, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/deepseek-ai/DeepSeek-OCR", "ollama": "https://ollama.com/library/deepseek-ocr", "mlx_4bit": "https://huggingface.co/mlx-community/DeepSeek-OCR-4bit" }, "run": { "ollama": "ollama run deepseek-ocr:3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/DeepSeek-OCR-4bit --prompt 'hello'" }, "description": "DeepSeek-OCR is a vision-language model that can perform token-efficient OCR.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "phi-4-14b", "name": "Phi-4 14B", "family": "Phi-4", "vendor": "Microsoft", "categories": [ "general" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-12-11", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 14.66, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/microsoft/phi-4/resolve/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 16384, "native_source": "HF config.json", "ollama_default": 16384, "ollama_source": "https://ollama.com/library/phi4" }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 9.1, "tag": "phi4:14b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 16.0, "tag": "phi4:14b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 29.0, "tag": "phi4:14b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 9.1, "tag": "phi4:14b", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 8.25, "repo": "mlx-community/phi-4-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/phi-4-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 9.1, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 12.5, "min_memory_gb_q8": 20.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Phi-4", "value": 5.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Phi-4", "value": 57.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Phi-4", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Phi-4", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Phi-4", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Phi-4", "value": 23.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Phi-4", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "phi-4", "value": 1256, "unit": "elo", "ci": 5, "votes": 24126, "rank": 320, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 5.9, "lmarena_elo": 1256 }, "popularity": { "ollama_pulls": "7.8M", "hf_downloads_30d": 363287, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/microsoft/phi-4", "ollama": "https://ollama.com/library/phi4", "mlx_4bit": "https://huggingface.co/mlx-community/phi-4-4bit" }, "run": { "ollama": "ollama run phi4:14b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/phi-4-4bit --prompt 'hello'" }, "description": "Phi-4 is a 14B parameter, state-of-the-art open model from Microsoft.", "notes": [], "gap": { "aa_points_behind_frontier": 51.7, "aa_pct_of_frontier": 10.2, "lmarena_elo_behind_frontier": 268, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "phi-4-mini", "name": "Phi-4-mini 3.8B", "family": "Phi-4", "vendor": "Microsoft", "categories": [ "edge" ], "status": "legacy", "open_weights": true, "release_date": "2025-02-19", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.84, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/phi4-mini" }, "inputs": "Text", "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 2.5, "tag": "phi4-mini:3.8b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-mini/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 4.1, "tag": "phi4-mini:3.8b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-mini/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 7.7, "tag": "phi4-mini:3.8b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-mini/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 2.5, "tag": "phi4-mini:3.8b", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-mini/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 2.16, "repo": "mlx-community/Phi-4-mini-instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Phi-4-mini-instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 2.5, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 4.9, "min_memory_gb_q8": 6.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Phi-4 Mini Instruct", "value": 6.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Phi-4 Mini Instruct", "value": 33.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Phi-4 Mini Instruct", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Phi-4 Mini Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Phi-4 Mini Instruct", "value": 0.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Phi-4 Mini Instruct", "value": 8.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Phi-4 Mini Instruct", "value": 21.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Phi-4 Mini Instruct", "value": 15.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/phi-4-mini", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 6.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.5M", "hf_downloads_30d": 358611, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "ollama": "https://ollama.com/library/phi4-mini", "mlx_4bit": "https://huggingface.co/mlx-community/Phi-4-mini-instruct-4bit" }, "run": { "ollama": "ollama run phi4-mini:3.8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Phi-4-mini-instruct-4bit --prompt 'hello'" }, "description": "Phi-4-mini brings significant enhancements in multilingual support, reasoning, and mathematics, and now, the long-awaited function calling feature is finally supported.", "notes": [], "gap": { "aa_points_behind_frontier": 51.3, "aa_pct_of_frontier": 10.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "phi-4-reasoning-plus", "name": "Phi-4-reasoning-plus 14B", "family": "Phi-4", "vendor": "Microsoft", "categories": [ "reasoning" ], "status": "legacy", "open_weights": true, "release_date": "2025-04-17", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 14.66, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus/resolve/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 32768, "native_source": "HF config.json", "ollama_default": 32768, "ollama_source": "https://ollama.com/library/phi4-reasoning" }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 11.0, "tag": "phi4-reasoning:14b-plus-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-reasoning/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 17.0, "tag": "phi4-reasoning:14b-plus-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-reasoning/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 29.0, "tag": "phi4-reasoning:14b-plus-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/phi4-reasoning/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 9.16, "repo": "mlx-community/Phi-4-reasoning-plus-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Phi-4-reasoning-plus-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 11.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 14.6, "min_memory_gb_q8": 21.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.7M", "hf_downloads_30d": 7251, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/microsoft/Phi-4-reasoning-plus", "ollama": "https://ollama.com/library/phi4-reasoning", "mlx_4bit": "https://huggingface.co/mlx-community/Phi-4-reasoning-plus-4bit" }, "run": { "ollama": "ollama run phi4-reasoning:14b-plus", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Phi-4-reasoning-plus-4bit --prompt 'hello'" }, "description": "Phi 4 reasoning and reasoning plus are 14-billion parameter open-weight reasoning models that rival much larger models on complex reasoning tasks. ", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "glm-5.3", "name": "GLM-5.3", "family": "GLM-5", "vendor": "Z.ai", "categories": [ "coding", "agentic", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-08-25", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 753.33, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "glm-5.3", "url": "https://huggingface.co/zai-org/GLM-5.3", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/glm-5.3" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q8_0": { "value_gb": 801.36, "repo": "unsloth/GLM-5.3-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/GLM-5.3-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 418.32, "repo": "mlx-community/GLM-5.3-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/GLM-5.3-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 418.32, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 483.1, "min_memory_gb_q8": 923.6, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "GLM-5.3 (Max)", "value": 44.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "GLM-5.3 (Max)", "value": 91.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-5.3 (Max)", "value": 42.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "GLM-5.3 (Max)", "value": 83.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "GLM-5.3 (Max)", "value": 59.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-5.3 (Max)", "value": 79.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "GLM-5.3 (Low)", "value": 34.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-5.3 (Low)", "value": 36.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "GLM-5.3 (Low)", "value": 42.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-5.3 (Low)", "value": 72.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2.1", "value": 88.2, "unit": "%", "column": "GLM-5.3", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/zai-org/GLM-5.3", "as_of": "2026-10-14", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "glm-5.3-max", "value": 1479, "unit": "elo", "ci": 5, "votes": 22543, "rank": 28, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 44.8, "lmarena_elo": 1479 }, "popularity": { "ollama_pulls": "96.7K", "hf_downloads_30d": 1803969, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/zai-org/GLM-5.3", "ollama": "https://ollama.com/library/glm-5.3", "mlx_4bit": "https://huggingface.co/mlx-community/GLM-5.3-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/GLM-5.3-4bit --prompt 'hello'" }, "description": "Z.ai's flagship model and the most capable open-weights model for coding, with major gains on long-horizon agentic tasks.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 12.8, "aa_pct_of_frontier": 77.8, "lmarena_elo_behind_frontier": 45, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "glm-5.3-flash", "name": "GLM-5.3-Flash", "family": "GLM-5", "vendor": "Z.ai", "categories": [ "coding", "agentic", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-08-25", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 321.32, "active_b": 18, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"320B total parameters and just 18B active parameters\" (https://huggingface.co/zai-org/GLM-5.3-Flash)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/glm-5.3-flash" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 200.83, "repo": "bartowski/GLM-5.3-Flash-BF16-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/GLM-5.3-Flash-BF16-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 340.98, "repo": "unsloth/GLM-5.3-Flash-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/GLM-5.3-Flash-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 203.99, "repo": "mlx-community/GLM-5.3-Flash-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/GLM-5.3-Flash-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 200.83, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 233.0, "min_memory_gb_q8": 394.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "GLM-5.3-Flash", "value": 41.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "GLM-5.3-Flash", "value": 91.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-5.3-Flash", "value": 39.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "GLM-5.3-Flash", "value": 84.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "GLM-5.3-Flash", "value": 51.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-5.3-Flash", "value": 80.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "glm-5.3-flash", "value": 1474, "unit": "elo", "ci": 5, "votes": 32925, "rank": 41, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 41.8, "lmarena_elo": 1474 }, "popularity": { "ollama_pulls": "185.5K", "hf_downloads_30d": 6531090, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/zai-org/GLM-5.3-Flash", "ollama": "https://ollama.com/library/glm-5.3-flash", "mlx_4bit": "https://huggingface.co/mlx-community/GLM-5.3-Flash-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/GLM-5.3-Flash-4bit --prompt 'hello'" }, "description": "Z.ai's first natively multimodal model, approaching Claude Opus 4.8 on coding and agentic benchmarks with just 18B active parameters.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows.", "llama.cpp supports it from v0.6.0 (released 2026-10-05, per the GitHub release notes)." ], "gap": { "aa_points_behind_frontier": 15.8, "aa_pct_of_frontier": 72.6, "lmarena_elo_behind_frontier": 50, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "glm-4.7-flash", "name": "GLM-4.7-Flash 30B-A3B", "family": "GLM-4", "vendor": "Z.ai", "categories": [ "general", "coding" ], "status": "legacy-popular", "open_weights": true, "release_date": "2026-01-19", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 31.22, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 202752, "native_source": "HF config.json", "ollama_default": 202752, "ollama_source": "https://ollama.com/library/glm-4.7-flash" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "glm-4.7-flash:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-4.7-flash/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 32.0, "tag": "glm-4.7-flash:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-4.7-flash/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 60.0, "tag": "glm-4.7-flash:bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-4.7-flash/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "glm-4.7-flash:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-4.7-flash/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.85, "repo": "mlx-community/GLM-4.7-Flash-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/GLM-4.7-Flash-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 38.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "GLM-4.7-Flash (Reasoning)", "value": 14.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "GLM-4.7-Flash (Reasoning)", "value": 58.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-4.7-Flash (Reasoning)", "value": 7.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "GLM-4.7-Flash (Reasoning)", "value": 22.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "GLM-4.7-Flash (Reasoning)", "value": 98.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "GLM-4.7-Flash (Reasoning)", "value": 60.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-4.7-Flash (Reasoning)", "value": 41.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 10.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 45.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 5.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 3.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 91.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 46.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-4.7-Flash (Non-reasoning)", "value": 20.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "glm-4.7-flash", "value": 1365, "unit": "elo", "ci": 6, "votes": 12021, "rank": 201, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 14.9, "lmarena_elo": 1365 }, "popularity": { "ollama_pulls": "2.1M", "hf_downloads_30d": 1307102, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/zai-org/GLM-4.7-Flash", "ollama": "https://ollama.com/library/glm-4.7-flash", "mlx_4bit": "https://huggingface.co/mlx-community/GLM-4.7-Flash-4bit" }, "run": { "ollama": "ollama run glm-4.7-flash", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/GLM-4.7-Flash-4bit --prompt 'hello'" }, "description": "As the strongest model in the 30B class, GLM-4.7-Flash offers a new option for lightweight deployment that balances performance and efficiency.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 42.7, "aa_pct_of_frontier": 25.9, "lmarena_elo_behind_frontier": 159, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "glm-4.5-air", "name": "GLM-4.5-Air 106B-A12B", "family": "GLM-4", "vendor": "Z.ai", "categories": [ "general", "coding" ], "status": "legacy", "open_weights": true, "release_date": "2025-07-20", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 110.47, "active_b": 12.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/zai-org/GLM-4.5-Air", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 72.98, "repo": "unsloth/GLM-4.5-Air-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/GLM-4.5-Air-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 117.46, "repo": "unsloth/GLM-4.5-Air-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/GLM-4.5-Air-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 60.14, "repo": "mlx-community/GLM-4.5-Air-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/GLM-4.5-Air-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 72.98, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 85.9, "min_memory_gb_q8": 137.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "GLM-4.5-Air", "value": 11.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "GLM-4.5-Air", "value": 73.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "GLM-4.5-Air", "value": 7.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "GLM-4.5-Air", "value": 20.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "GLM-4.5-Air", "value": 46.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "GLM-4.5-Air", "value": 37.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "GLM-4.5-Air", "value": 46.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/glm-4-5-air", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "glm-4.5-air", "value": 1373, "unit": "elo", "ci": 4, "votes": 30660, "rank": 196, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 11.1, "lmarena_elo": 1373 }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 701853, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/zai-org/GLM-4.5-Air", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/GLM-4.5-Air-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/GLM-4.5-Air-4bit --prompt 'hello'" }, "description": null, "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 46.5, "aa_pct_of_frontier": 19.3, "lmarena_elo_behind_frontier": 151, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "glm-ocr", "name": "GLM-OCR", "family": "GLM-OCR", "vendor": "Z.ai", "categories": [ "vision", "ocr" ], "status": "current", "open_weights": true, "release_date": "2026-01-30", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1.33, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/zai-org/GLM-OCR", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/glm-ocr" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools" ], "files": { "q8_0": { "value_gb": 1.8, "tag": "glm-ocr:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-ocr/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 2.7, "tag": "glm-ocr:bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-ocr/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 2.7, "tag": "glm-ocr:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/glm-ocr/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 2.7, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 5.1, "min_memory_gb_q8": 4.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "8.3M", "hf_downloads_30d": 1703885, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/zai-org/GLM-OCR", "ollama": "https://ollama.com/library/glm-ocr", "mlx_4bit": null }, "run": { "ollama": "ollama run glm-ocr", "ollama_cloud_only": false, "mlx": null }, "description": "GLM-OCR is a multimodal OCR model for complex document understanding, built on the GLM-V encoder–decoder architecture.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "kimi-k3", "name": "Kimi K3", "family": "Kimi", "vendor": "Moonshot AI", "categories": [ "general", "agentic", "vision", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-06-13", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 2779.93, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "kimi-k3", "url": "https://huggingface.co/moonshotai/Kimi-K3", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/kimi-k3" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": {}, "memory": { "weights_gb_4bit": 1667.96, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 1920.2, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Kimi K3 (Max)", "value": 43.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Kimi K3 (Max)", "value": 93.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi K3 (Max)", "value": 46.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Kimi K3 (Max)", "value": 85.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Kimi K3 (Max)", "value": 59.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi K3 (Max)", "value": 88.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Kimi K3 (Max)", "value": 80.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Kimi K3 (Low)", "value": 30.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Kimi K3 (Low)", "value": 84.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi K3 (Low)", "value": 25.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Kimi K3 (Low)", "value": 82.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Kimi K3 (Low)", "value": 52.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi K3 (Low)", "value": 79.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Kimi K3 (Low)", "value": 78.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k3-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 93.5, "unit": "%", "column": "Kimi K3 (max)", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K3", "as_of": "2026-09-02", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.1", "value": 88.3, "unit": "%", "column": "Kimi K3 (max)", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K3", "as_of": "2026-09-02", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "kimi-k3-max", "value": 1487, "unit": "elo", "ci": 5, "votes": 32795, "rank": 18, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 43.6, "lmarena_elo": 1487 }, "popularity": { "ollama_pulls": "104.1K", "hf_downloads_30d": 1063166, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/moonshotai/Kimi-K3", "ollama": "https://ollama.com/library/kimi-k3", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date.", "notes": [ "~2.8T total params (HF safetensors); local use realistically needs a multi-node or 1.5TB+ setup.", "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 14.0, "aa_pct_of_frontier": 75.7, "lmarena_elo_behind_frontier": 37, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "kimi-k2.7-code", "name": "Kimi K2.7 Code", "family": "Kimi", "vendor": "Moonshot AI", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-06-11", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1026.88, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "modified-mit", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/kimi-k2.7-code" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": { "mlx_4bit": { "value_gb": 641.01, "repo": "mlx-community/Kimi-K2.7-Code-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Kimi-K2.7-Code-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 641.01, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 739.2, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Kimi K2.7 Code", "value": 25.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Kimi K2.7 Code", "value": 89.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi K2.7 Code", "value": 35.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Kimi K2.7 Code", "value": 44.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Kimi K2.7 Code", "value": 67.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Kimi K2.7 Code", "value": 90.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Kimi K2.7 Code", "value": 47.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Kimi K2.7 Code", "value": 63.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi K2.7 Code", "value": 79.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-7-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 25.8, "lmarena_elo": null }, "popularity": { "ollama_pulls": "250.7K", "hf_downloads_30d": 113845, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "ollama": "https://ollama.com/library/kimi-k2.7-code", "mlx_4bit": "https://huggingface.co/mlx-community/Kimi-K2.7-Code-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/Kimi-K2.7-Code-4bit --prompt 'hello'" }, "description": "Kimi K2.7 Code is Moonshot AI's coding-focused agentic model built upon Kimi K2.6, with substantial improvements on real-world long-horizon coding tasks and roughly 30% lower thinking-token usage.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 31.8, "aa_pct_of_frontier": 44.8, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "kimi-k2.6", "name": "Kimi K2.6", "family": "Kimi", "vendor": "Moonshot AI", "categories": [ "general", "agentic", "vision" ], "status": "superseded", "open_weights": true, "release_date": "2026-04-14", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1026.88, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "modified-mit", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/kimi-k2.6" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": {}, "memory": { "weights_gb_4bit": 616.13, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 710.5, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Kimi K2.6 (Reasoning)", "value": 27.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Kimi K2.6 (Reasoning)", "value": 91.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi K2.6 (Reasoning)", "value": 37.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Kimi K2.6 (Reasoning)", "value": 43.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Kimi K2.6 (Reasoning)", "value": 65.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Kimi K2.6 (Reasoning)", "value": 95.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Kimi K2.6 (Reasoning)", "value": 51.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Kimi K2.6 (Reasoning)", "value": 76.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi K2.6 (Reasoning)", "value": 81.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "MMMU-Pro", "variant": "Kimi K2.6 (Reasoning)", "value": 79.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA Intelligence Index", "variant": "Kimi K2.6 (Non-reasoning)", "value": 23.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Kimi K2.6 (Non-reasoning)", "value": 78.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi K2.6 (Non-reasoning)", "value": 19.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Kimi K2.6 (Non-reasoning)", "value": 37.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Kimi K2.6 (Non-reasoning)", "value": 93.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Kimi K2.6 (Non-reasoning)", "value": 44.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi K2.6 (Non-reasoning)", "value": 69.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.0 (Terminus-2)", "value": 66.7, "unit": "%", "column": "Kimi K2.6", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "as_of": "2026-05-19", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-Bench Verified", "value": 80.2, "unit": "%", "column": "Kimi K2.6", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "as_of": "2026-05-19", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench (v6)", "value": 89.6, "unit": "%", "column": "Kimi K2.6", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "as_of": "2026-05-19", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA-Diamond", "value": 90.5, "unit": "%", "column": "Kimi K2.6", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "as_of": "2026-05-19", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "kimi-k2.6", "value": 1461, "unit": "elo", "ci": 4, "votes": 40013, "rank": 59, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 27.0, "lmarena_elo": 1461 }, "popularity": { "ollama_pulls": "491.5K", "hf_downloads_30d": 396994, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/moonshotai/Kimi-K2.6", "ollama": "https://ollama.com/library/kimi-k2.6", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "Kimi K2.6 is an open-source, native multimodal agentic model that advances practical capabilities in long-horizon coding, coding-driven design, proactive autonomous execution, and swarm-based task orchestration.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows.", "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 30.6, "aa_pct_of_frontier": 46.9, "lmarena_elo_behind_frontier": 63, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "kimi-linear-48b-a3b", "name": "Kimi Linear 48B-A3B", "family": "Kimi", "vendor": "Moonshot AI", "categories": [ "general", "long-context" ], "status": "legacy", "open_weights": true, "release_date": "2025-10-30", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 49.12, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/moonshotai/Kimi-Linear-48B-A3B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 30.06, "repo": "bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 52.25, "repo": "bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 27.64, "repo": "mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 30.06, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 36.6, "min_memory_gb_q8": 62.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Kimi Linear 48B A3B Instruct", "value": 7.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Kimi Linear 48B A3B Instruct", "value": 41.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Kimi Linear 48B A3B Instruct", "value": 2.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Kimi Linear 48B A3B Instruct", "value": 11.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Kimi Linear 48B A3B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Kimi Linear 48B A3B Instruct", "value": 28.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Kimi Linear 48B A3B Instruct", "value": 28.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 7.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 197567, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/moonshotai/Kimi-Linear-48B-A3B-Instruct", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 50.3, "aa_pct_of_frontier": 12.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gpt-oss-20b", "name": "gpt-oss-20b", "family": "gpt-oss", "vendor": "OpenAI", "categories": [ "general", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2025-08-04", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 20.91, "active_b": 3.6, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"21B parameters with 3.6B active parameters\" (https://huggingface.co/openai/gpt-oss-20b)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/openai/gpt-oss-20b", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gpt-oss" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "ollama_default": { "value_gb": 14.0, "tag": "gpt-oss:20b", "source_name": "Ollama library", "url": "https://ollama.com/library/gpt-oss/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 11.62, "repo": "unsloth/gpt-oss-20b-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/gpt-oss-20b-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 11.62, "repo": "mlx-community/gpt-oss-20b-OptiQ-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gpt-oss-20b-OptiQ-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 11.62, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 15.4, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "gpt-oss-20b (High)", "value": 9.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "gpt-oss-20b (High)", "value": 68.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "gpt-oss-20b (High)", "value": 11.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "gpt-oss-20b (High)", "value": 10.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "gpt-oss-20b (High)", "value": 13.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "gpt-oss-20b (High)", "value": 60.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "gpt-oss-20b (High)", "value": 38.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "gpt-oss-20b (High)", "value": 65.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "gpt-oss-20b (High)", "value": 34.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "gpt-oss-20b (Low)", "value": 10.0, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "gpt-oss-20b (Low)", "value": 61.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "gpt-oss-20b (Low)", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "gpt-oss-20b (Low)", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "gpt-oss-20b (Low)", "value": 50.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "gpt-oss-20b (Low)", "value": 57.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "gpt-oss-20b (Low)", "value": 31.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-20b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "gpt-oss-20b", "value": 1317, "unit": "elo", "ci": 6, "votes": 10419, "rank": 267, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 10.0, "lmarena_elo": 1317 }, "popularity": { "ollama_pulls": "13.8M", "hf_downloads_30d": 5988822, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/openai/gpt-oss-20b", "ollama": "https://ollama.com/library/gpt-oss", "mlx_4bit": "https://huggingface.co/mlx-community/gpt-oss-20b-OptiQ-4bit" }, "run": { "ollama": "ollama run gpt-oss:20b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gpt-oss-20b-OptiQ-4bit --prompt 'hello'" }, "description": "OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases.", "notes": [ "Native MXFP4 weights; the Q4_K_M file listed is from a community GGUF repo and is close to the native size." ], "gap": { "aa_points_behind_frontier": 47.6, "aa_pct_of_frontier": 17.4, "lmarena_elo_behind_frontier": 207, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "gpt-oss-120b", "name": "gpt-oss-120b", "family": "gpt-oss", "vendor": "OpenAI", "categories": [ "general", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2025-08-04", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 116.83, "active_b": 5.1, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"117B parameters with 5.1B active parameters\" (https://huggingface.co/openai/gpt-oss-120b)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/openai/gpt-oss-120b", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/gpt-oss" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "ollama_default": { "value_gb": 65.0, "tag": "gpt-oss:120b", "source_name": "Ollama library", "url": "https://ollama.com/library/gpt-oss/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 62.77, "repo": "unsloth/gpt-oss-120b-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/gpt-oss-120b-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 65.77, "repo": "mlx-community/gpt-oss-120b-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/gpt-oss-120b-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 62.77, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 74.2, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "gpt-oss-120b (High)", "value": 11.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "gpt-oss-120b (High)", "value": 78.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "gpt-oss-120b (High)", "value": 19.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "gpt-oss-120b (High)", "value": 23.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "gpt-oss-120b (High)", "value": 26.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "gpt-oss-120b (High)", "value": 65.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "gpt-oss-120b (High)", "value": 34.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "gpt-oss-120b (High)", "value": 69.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "gpt-oss-120b (High)", "value": 52.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "gpt-oss-120b (Low)", "value": 10.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "gpt-oss-120b (Low)", "value": 67.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "gpt-oss-120b (Low)", "value": 5.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "gpt-oss-120b (Low)", "value": 5.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "gpt-oss-120b (Low)", "value": 13.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "gpt-oss-120b (Low)", "value": 45.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "gpt-oss-120b (Low)", "value": 58.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "gpt-oss-120b (Low)", "value": 46.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/gpt-oss-120b-low", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "gpt-oss-120b", "value": 1352, "unit": "elo", "ci": 4, "votes": 30206, "rank": 215, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 11.6, "lmarena_elo": 1352 }, "popularity": { "ollama_pulls": "13.8M", "hf_downloads_30d": 3423004, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/openai/gpt-oss-120b", "ollama": "https://ollama.com/library/gpt-oss", "mlx_4bit": "https://huggingface.co/mlx-community/gpt-oss-120b-4bit" }, "run": { "ollama": "ollama run gpt-oss:120b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/gpt-oss-120b-4bit --prompt 'hello'" }, "description": "OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases.", "notes": [ "Native MXFP4 weights (Ollama default 65GB)." ], "gap": { "aa_points_behind_frontier": 46.0, "aa_pct_of_frontier": 20.1, "lmarena_elo_behind_frontier": 172, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "granite-4.2-3b", "name": "Granite 4.2 3B", "family": "Granite 4", "vendor": "IBM", "categories": [ "edge", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2026-08-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.66, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/ibm-granite/granite-4.2-3b", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/granite4.2" }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 2.2, "tag": "granite4.2:3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 3.9, "tag": "granite4.2:3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 7.3, "tag": "granite4.2:3b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 2.2, "tag": "granite4.2:3b", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 2.06, "repo": "lmstudio-community/granite-4.2-3b-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/lmstudio-community/granite-4.2-3b-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 2.2, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 4.5, "min_memory_gb_q8": 6.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Granite 4.2 3B", "value": 9.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Granite 4.2 3B", "value": 55.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Granite 4.2 3B", "value": 6.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Granite 4.2 3B", "value": 13.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Granite 4.2 3B", "value": 25.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Granite 4.2 3B", "value": 24.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "granite-4.2-3b", "value": 1290, "unit": "elo", "ci": 12, "votes": 3251, "rank": 292, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 9.1, "lmarena_elo": 1290 }, "popularity": { "ollama_pulls": "144.8K", "hf_downloads_30d": 42635, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ibm-granite/granite-4.2-3b", "ollama": "https://ollama.com/library/granite4.2", "mlx_4bit": "https://huggingface.co/lmstudio-community/granite-4.2-3b-MLX-4bit" }, "run": { "ollama": "ollama run granite4.2:3b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model lmstudio-community/granite-4.2-3b-MLX-4bit --prompt 'hello'" }, "description": "IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, thinking and structured JSON output. Released under Apache 2.0 license. ", "notes": [], "gap": { "aa_points_behind_frontier": 48.5, "aa_pct_of_frontier": 15.8, "lmarena_elo_behind_frontier": 234, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "granite-4.2-8b", "name": "Granite 4.2 8B", "family": "Granite 4", "vendor": "IBM", "categories": [ "general", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2026-08-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.79, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/ibm-granite/granite-4.2-8b", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/granite4.2" }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 5.3, "tag": "granite4.2:8b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.3, "tag": "granite4.2:8b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 18.0, "tag": "granite4.2:8b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 5.3, "tag": "granite4.2:8b", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 4.95, "repo": "lmstudio-community/granite-4.2-8b-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/lmstudio-community/granite-4.2-8b-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.3, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.1, "min_memory_gb_q8": 12.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Granite 4.2 8B", "value": 11.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Granite 4.2 8B", "value": 63.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Granite 4.2 8B", "value": 9.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Granite 4.2 8B", "value": 18.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Granite 4.2 8B", "value": 31.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Granite 4.2 8B", "value": 45.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-8b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "granite-4.2-8b", "value": 1290, "unit": "elo", "ci": 11, "votes": 3254, "rank": 290, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 11.1, "lmarena_elo": 1290 }, "popularity": { "ollama_pulls": "144.8K", "hf_downloads_30d": 105751, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ibm-granite/granite-4.2-8b", "ollama": "https://ollama.com/library/granite4.2", "mlx_4bit": "https://huggingface.co/lmstudio-community/granite-4.2-8b-MLX-4bit" }, "run": { "ollama": "ollama run granite4.2:8b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model lmstudio-community/granite-4.2-8b-MLX-4bit --prompt 'hello'" }, "description": "IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, thinking and structured JSON output. Released under Apache 2.0 license. ", "notes": [], "gap": { "aa_points_behind_frontier": 46.5, "aa_pct_of_frontier": 19.3, "lmarena_elo_behind_frontier": 234, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "granite-4.2-30b", "name": "Granite 4.2 30B", "family": "Granite 4", "vendor": "IBM", "categories": [ "general", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2026-08-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 29.28, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/ibm-granite/granite-4.2-30b", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/granite4.2" }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 18.0, "tag": "granite4.2:30b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 31.0, "tag": "granite4.2:30b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 45.0, "tag": "granite4.2:30b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 18.0, "tag": "granite4.2:30b", "source_name": "Ollama library", "url": "https://ollama.com/library/granite4.2/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 16.47, "repo": "lmstudio-community/granite-4.2-30b-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/lmstudio-community/granite-4.2-30b-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 18.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 22.7, "min_memory_gb_q8": 37.6, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Granite 4.2 30B", "value": 12.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Granite 4.2 30B", "value": 64.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Granite 4.2 30B", "value": 11.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Granite 4.2 30B", "value": 26.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Granite 4.2 30B", "value": 37.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Granite 4.2 30B", "value": 49.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/granite-4-2-30b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "granite-4.2-30b", "value": 1340, "unit": "elo", "ci": 10, "votes": 3401, "rank": 235, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 12.7, "lmarena_elo": 1340 }, "popularity": { "ollama_pulls": "144.8K", "hf_downloads_30d": 37104, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ibm-granite/granite-4.2-30b", "ollama": "https://ollama.com/library/granite4.2", "mlx_4bit": "https://huggingface.co/lmstudio-community/granite-4.2-30b-MLX-4bit" }, "run": { "ollama": "ollama run granite4.2:30b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model lmstudio-community/granite-4.2-30b-MLX-4bit --prompt 'hello'" }, "description": "IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, thinking and structured JSON output. Released under Apache 2.0 license. ", "notes": [], "gap": { "aa_points_behind_frontier": 44.9, "aa_pct_of_frontier": 22.0, "lmarena_elo_behind_frontier": 184, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "granite-embedding-278m", "name": "Granite Embedding 278M", "family": "Granite Embedding", "vendor": "IBM", "categories": [ "embedding" ], "status": "legacy", "open_weights": true, "release_date": "2024-12-04", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.28, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/ibm-granite/granite-embedding-278m-multilingual", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 514, "native_source": "HF config.json", "ollama_default": 512, "ollama_source": "https://ollama.com/library/granite-embedding" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "bf16": { "value_gb": 0.563, "tag": "granite-embedding:278m-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/granite-embedding/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.563, "tag": "granite-embedding:278m", "source_name": "Ollama library", "url": "https://ollama.com/library/granite-embedding/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 0.22, "repo": "lmstudio-community/granite-embedding-278m-multilingual-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/lmstudio-community/granite-embedding-278m-multilingual-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 0.3, "repo": "lmstudio-community/granite-embedding-278m-multilingual-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/lmstudio-community/granite-embedding-278m-multilingual-GGUF", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.22, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 2.3, "min_memory_gb_q8": 2.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "389.1K", "hf_downloads_30d": 63894, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ibm-granite/granite-embedding-278m-multilingual", "ollama": "https://ollama.com/library/granite-embedding", "mlx_4bit": null }, "run": { "ollama": "ollama run granite-embedding:278m", "ollama_cloud_only": false, "mlx": null }, "description": "The IBM Granite Embedding 30M and 278M models models are text-only dense biencoder embedding models, with 30M available in English only and 278M serving multilingual use cases.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3.5-super-120b", "name": "Nemotron 3.5 Super VL 120B-A12B", "family": "Nemotron 3.5", "vendor": "NVIDIA", "categories": [ "general", "vision", "coding", "agentic", "tool-calling", "long-context", "ocr" ], "status": "current", "open_weights": true, "release_date": "2026-10-12", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 124.45, "active_b": 12.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "openmdw-1.1", "url": "https://openmdw.ai/license/1-1/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/nemotron-3.5-super" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 90.0, "tag": "nemotron-3.5-super:120b-a12b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 133.0, "tag": "nemotron-3.5-super:120b-a12b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 249.0, "tag": "nemotron-3.5-super:120b-a12b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 80.0, "tag": "nemotron-3.5-super:120b-a12b-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 130.0, "tag": "nemotron-3.5-super:120b-a12b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 90.0, "tag": "nemotron-3.5-super:120b-a12b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-super/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 90.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 105.5, "min_memory_gb_q8": 154.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3.5 Super VL", "value": 26.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-super-vl", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3.5 Super VL", "value": 30.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-super-vl", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Nemotron 3.5 Super VL", "value": 47.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-super-vl", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3.5 Super VL", "value": 69.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-super-vl", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "nvidia-nemotron-3.5-super-vl-120b-a12b", "value": 1416, "unit": "elo", "ci": 12, "votes": 2523, "rank": 138, "preliminary": true, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 26.5, "lmarena_elo": 1416 }, "popularity": { "ollama_pulls": "1.8K", "hf_downloads_30d": 2155, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16", "ollama": "https://ollama.com/library/nemotron-3.5-super", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3.5-super:120b-a12b", "ollama_cloud_only": false, "mlx": null }, "description": "NVIDIA's 120B open model (12B active per word) that reads text and pictures, can think first, calls tools and takes up to 1M tokens.", "notes": [ "Added to Ollama in v0.40.5 (2026-10-15).", "Ollama's 4-bit build is 80-90 GB, so it needs a 128 GB machine.", "NVIDIA released weights, training data and recipes (OpenMDW 1.1 licence).", "Usable context depends on your free memory, not the 1M maximum." ], "gap": { "aa_points_behind_frontier": 31.1, "aa_pct_of_frontier": 46.0, "lmarena_elo_behind_frontier": 108, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3.5-bolt-74b", "name": "Nemotron 3.5 Bolt 74B-A9B", "family": "Nemotron 3.5", "vendor": "NVIDIA", "categories": [ "general", "vision", "coding", "agentic", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2026-10-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 74, "active_b": 9, "is_moe": true, "total_source": "model card: \"74B Total / 9B Active\" (HF repo is a 24GB compressed build)", "active_source": "\"74B Total / 9B Active\" (https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB)" }, "license": { "id": "other", "name": "openmdw-1.1", "url": "https://openmdw.ai/license/1-1/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/nemotron-3.5-bolt" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "ollama_default": { "value_gb": 25.0, "tag": "nemotron-3.5-bolt:74b-a9b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-bolt/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 25.0, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 30.7, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3.5 Bolt 74B-A9B", "value": 21.4, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-bolt-74b-a9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3.5 Bolt 74B-A9B", "value": 24.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-bolt-74b-a9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Nemotron 3.5 Bolt 74B-A9B", "value": 39.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-bolt-74b-a9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3.5 Bolt 74B-A9B", "value": 69.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-bolt-74b-a9b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.1", "value": 50.98, "unit": "%", "column": "Nemotron 3.5 Bolt 74B-A9B-24GB (IQ2)", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB", "as_of": "2026-10-15", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 86.43, "unit": "%", "column": "Nemotron 3.5 Bolt 74B-A9B-24GB (IQ2)", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 21.4, "lmarena_elo": null }, "popularity": { "ollama_pulls": "118", "hf_downloads_30d": 58, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB", "ollama": "https://ollama.com/library/nemotron-3.5-bolt", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3.5-bolt:74b-a9b", "ollama_cloud_only": false, "mlx": null }, "description": "Nemotron 3.5 Bolt is NVIDIA’s compact vision-language model for local coding, reasoning, and tool use.", "notes": [ "Official NVIDIA \"24GB\" build of a 74B-A9B model; Ollama ships it as an IQ2 quant (25GB). Quality at this compression is untested by us.", "NVIDIA blog (2026-10-13): built for local AI agents, and fully open (training data, weights and recipes). A tiny 4GB sibling, Bolt E4B, is due 2026-10-23." ], "gap": { "aa_points_behind_frontier": 36.2, "aa_pct_of_frontier": 37.2, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3.5-lightning-30b", "name": "Nemotron 3.5 Lightning 30B-A3B", "family": "Nemotron 3.5", "vendor": "NVIDIA", "categories": [ "general", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-08-01", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 31.58, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "openmdw-1.1", "url": "https://openmdw.ai/license/1-1/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/nemotron-3.5-lightning" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 25.0, "tag": "nemotron-3.5-lightning:30b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 35.0, "tag": "nemotron-3.5-lightning:30b-a3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 66.0, "tag": "nemotron-3.5-lightning:30b-a3b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 23.0, "tag": "nemotron-3.5-lightning:30b-a3b-mlx", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 34.0, "tag": "nemotron-3.5-lightning:30b-a3b-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 25.0, "tag": "nemotron-3.5-lightning:30b-a3b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3.5-lightning/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 25.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 30.7, "min_memory_gb_q8": 42.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3.5 Lightning", "value": 12.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Nemotron 3.5 Lightning", "value": 74.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3.5 Lightning", "value": 10.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Nemotron 3.5 Lightning", "value": 24.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Nemotron 3.5 Lightning", "value": 32.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3.5 Lightning", "value": 60.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-5-lightning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond (no tools)", "value": 75.44, "unit": "%", "column": "Nemotron-3.5-Lightning-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 51.56, "unit": "%", "column": "Nemotron-3.5-Lightning-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.1", "value": 24.58, "unit": "%", "column": "Nemotron-3.5-Lightning-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "nvidia-nemotron-3.5-lightning-30b-a3b-nvfp4", "value": 1351, "unit": "elo", "ci": 6, "votes": 19938, "rank": 216, "preliminary": true, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 12.9, "lmarena_elo": 1351 }, "popularity": { "ollama_pulls": "234.9K", "hf_downloads_30d": 676533, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "ollama": "https://ollama.com/library/nemotron-3.5-lightning", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3.5-lightning:30b-a3b", "ollama_cloud_only": false, "mlx": null }, "description": "NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts (MoE) model with 3B active parameters built for always-on agents.", "notes": [], "gap": { "aa_points_behind_frontier": 44.7, "aa_pct_of_frontier": 22.4, "lmarena_elo_behind_frontier": 173, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3-super-120b", "name": "Nemotron 3 Super 120B-A12B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "general", "agentic" ], "status": "superseded", "open_weights": true, "release_date": "2026-03-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 123.61, "active_b": 12.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "nvidia-nemotron-open-model-license", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/nemotron-3-super" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 87.0, "tag": "nemotron-3-super:120b-a12b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-super/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 132.0, "tag": "nemotron-3-super:120b-a12b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-super/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 247.0, "tag": "nemotron-3-super:120b-a12b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-super/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 87.0, "tag": "nemotron-3-super:120b-a12b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-super/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 87.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 102.0, "min_memory_gb_q8": 153.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 12.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 80.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 20.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 28.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 38.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 67.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 36.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 71.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3 Super 120B A12B (Reasoning)", "value": 65.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "nvidia-nemotron-3-super-120b-a12b", "value": 1360, "unit": "elo", "ci": 7, "votes": 7878, "rank": 206, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 12.8, "lmarena_elo": 1360 }, "popularity": { "ollama_pulls": "3M", "hf_downloads_30d": 1045001, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "ollama": "https://ollama.com/library/nemotron-3-super", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3-super:120b-a12b", "ollama_cloud_only": false, "mlx": null }, "description": "NVIDIA Nemotron 3 Super is a 120B open MoE model activating just 12B parameters to deliver maximum compute efficiency and accuracy for complex multi-agent applications.", "notes": [ "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 44.8, "aa_pct_of_frontier": 22.2, "lmarena_elo_behind_frontier": 164, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3-nano-30b", "name": "Nemotron 3 Nano 30B-A3B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "general" ], "status": "superseded", "open_weights": true, "release_date": "2025-12-04", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 31.58, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "nvidia-nemotron-open-model-license", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/nemotron-3-nano" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 24.0, "tag": "nemotron-3-nano:30b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 34.0, "tag": "nemotron-3-nano:30b-a3b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 63.0, "tag": "nemotron-3-nano:30b-a3b-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 24.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 29.6, "min_memory_gb_q8": 41.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 8.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 75.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 11.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 13.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 6.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 40.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 30.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 71.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)", "value": 38.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 6.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 39.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 4.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 12.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 25.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 37.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)", "value": 10.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 78.3, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "AIME 2025", "variant": "AIME25 (no tools)", "value": 89.1, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA (no tools)", "value": 73.0, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench (v6 2025-08–2025-05)", "value": 68.3, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench (hard subset)", "value": 8.5, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "nvidia-nemotron-3-nano-30b-a3b-bf16", "value": 1314, "unit": "elo", "ci": 5, "votes": 15770, "rank": 271, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 8.9, "lmarena_elo": 1314 }, "popularity": { "ollama_pulls": "910.9K", "hf_downloads_30d": 1093063, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "ollama": "https://ollama.com/library/nemotron-3-nano", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3-nano:30b-a3b", "ollama_cloud_only": false, "mlx": null }, "description": "Nemotron-3-Nano is a new Standard for Efficient, Open, and Intelligent Agentic Models, now updated with a 4B parameter count model.", "notes": [], "gap": { "aa_points_behind_frontier": 48.7, "aa_pct_of_frontier": 15.5, "lmarena_elo_behind_frontier": 210, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3-nano-4b", "name": "Nemotron 3 Nano 4B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "edge" ], "status": "current", "open_weights": true, "release_date": "2026-03-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.97, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "nvidia-nemotron-open-model-license", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/nemotron-3-nano" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q8_0": { "value_gb": 4.2, "tag": "nemotron-3-nano:4b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 8.0, "tag": "nemotron-3-nano:4b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 2.8, "tag": "nemotron-3-nano:4b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-3-nano/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 2.8, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 5.2, "min_memory_gb_q8": 6.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 4.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 51.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 4.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 6.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 3.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 28.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 16.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 58.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "NVIDIA Nemotron 3 Nano 4B", "value": 21.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AIME 2025", "variant": "AIME25", "value": 78.5, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-4B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "as_of": "2026-03-20", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 53.2, "unit": "%", "column": "NVIDIA-Nemotron-3-Nano-4B-BF16", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "as_of": "2026-03-20", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 4.8, "lmarena_elo": null }, "popularity": { "ollama_pulls": "910.9K", "hf_downloads_30d": 1566195, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "ollama": "https://ollama.com/library/nemotron-3-nano", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-3-nano:4b", "ollama_cloud_only": false, "mlx": null }, "description": "Nemotron-3-Nano is a new Standard for Efficient, Open, and Intelligent Agentic Models, now updated with a 4B parameter count model.", "notes": [], "gap": { "aa_points_behind_frontier": 52.8, "aa_pct_of_frontier": 8.3, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-cascade-2-30b", "name": "Nemotron Cascade 2 30B-A3B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "reasoning" ], "status": "current", "open_weights": true, "release_date": null, "release_date_basis": null, "params": { "total_b": 30.0, "active_b": 3.0, "is_moe": true, "total_source": "parsed from model name", "active_source": "parsed from model name (AB)" }, "license": { "id": null, "name": null, "url": null, "gated": false, "source": null }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 262144, "ollama_source": "https://ollama.com/library/nemotron-cascade-2" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 24.0, "tag": "nemotron-cascade-2:30b-a3b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron-cascade-2/tags", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 24.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 29.6, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron Cascade 2 30B A3B", "value": 11.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Nemotron Cascade 2 30B A3B", "value": 75.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron Cascade 2 30B A3B", "value": 12.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Nemotron Cascade 2 30B A3B", "value": 21.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Nemotron Cascade 2 30B A3B", "value": 20.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Nemotron Cascade 2 30B A3B", "value": 53.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Nemotron Cascade 2 30B A3B", "value": 80.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron Cascade 2 30B A3B", "value": 39.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-cascade-2-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 11.7, "lmarena_elo": null }, "popularity": { "ollama_pulls": "154.3K", "hf_downloads_30d": null, "retrieved": "2026-10-15" }, "links": { "huggingface": null, "ollama": "https://ollama.com/library/nemotron-cascade-2", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron-cascade-2:30b-a3b", "ollama_cloud_only": false, "mlx": null }, "description": "An open 30B MoE model from NVIDIA with 3B activated parameters that delivers strong reasoning and agentic capabilities.", "notes": [], "gap": { "aa_points_behind_frontier": 45.9, "aa_pct_of_frontier": 20.3, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3-ultra-550b", "name": "Nemotron 3 Ultra 550B-A55B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "general", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-06-03", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 560.52, "active_b": 55.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "openmdw-1.1", "url": "https://openmdw.ai/license/1-1/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/nemotron-3-ultra" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": {}, "memory": { "weights_gb_4bit": 336.31, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 388.8, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 22.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 86.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 28.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 36.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 53.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 83.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 40.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 81.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3 Ultra 550B A55B (Reasoning)", "value": 79.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 2.1", "value": 56.4, "unit": "%", "column": "N-3-Ultra
550B-A55B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-Bench Verified", "value": 70.7, "unit": "%", "column": "N-3-Ultra
550B-A55B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench (v6)", "value": 89.0, "unit": "%", "column": "N-3-Ultra
550B-A55B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA (no tools)", "value": 87.0, "unit": "%", "column": "N-3-Ultra
550B-A55B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 86.8, "unit": "%", "column": "N-3-Ultra
550B-A55B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "nvidia-nemotron-3-ultra-550b-a55b-nvfp4", "value": 1426, "unit": "elo", "ci": 7, "votes": 11567, "rank": 121, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 22.9, "lmarena_elo": 1426 }, "popularity": { "ollama_pulls": "126.1K", "hf_downloads_30d": 929843, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "ollama": "https://ollama.com/library/nemotron-3-ultra", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "NVIDIA Nemotron 3 Ultra is built for high-throughput reasoning and long-running agent workflows.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 34.7, "aa_pct_of_frontier": 39.8, "lmarena_elo_behind_frontier": 98, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nemotron-3-nano-omni-30b", "name": "Nemotron 3 Nano Omni 30B-A3B", "family": "Nemotron 3", "vendor": "NVIDIA", "categories": [ "vision", "audio", "video" ], "status": "current", "open_weights": true, "release_date": "2026-04-20", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 33.02, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "nvidia-open-model-agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/nemotron3" }, "inputs": "Text, Image", "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 28.0, "tag": "nemotron3:33b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 36.0, "tag": "nemotron3:33b-q8", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 66.0, "tag": "nemotron3:33b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 28.0, "tag": "nemotron3:33b", "source_name": "Ollama library", "url": "https://ollama.com/library/nemotron3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 28.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 34.2, "min_memory_gb_q8": 43.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 10.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 46.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 4.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 8.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 6.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 45.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 63.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 39.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "Nemotron 3 Nano Omni 30B A3B Reasoning", "value": 53.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 10.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": "680.3K", "hf_downloads_30d": 336466, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", "ollama": "https://ollama.com/library/nemotron3", "mlx_4bit": null }, "run": { "ollama": "ollama run nemotron3:33b", "ollama_cloud_only": false, "mlx": null }, "description": "NVIDIA Nemotron 3 Nano Omni is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows. ", "notes": [], "gap": { "aa_points_behind_frontier": 47.3, "aa_pct_of_frontier": 17.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "olmo-3.1-32b-think", "name": "OLMo 3.1 32B Think", "family": "OLMo 3", "vendor": "Ai2", "categories": [ "reasoning", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2025-12-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 32.23, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 65536, "native_source": "HF config.json", "ollama_default": 65536, "ollama_source": "https://ollama.com/library/olmo-3.1" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "olmo-3.1:32b-think-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 34.0, "tag": "olmo-3.1:32b-think-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 64.0, "tag": "olmo-3.1:32b-think-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "olmo-3.1:32b-think", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.13, "repo": "mlx-community/Olmo-3.1-32B-Think-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Olmo-3.1-32B-Think-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 41.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Olmo 3.1 32B Think", "value": 7.1, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Olmo 3.1 32B Think", "value": 59.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Olmo 3.1 32B Think", "value": 6.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Olmo 3.1 32B Think", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Olmo 3.1 32B Think", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Olmo 3.1 32B Think", "value": 66.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Olmo 3.1 32B Think", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AIME 2025", "variant": "AIME 2025", "value": 78.1, "unit": "%", "column": "Olmo 3.1 32B Think", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "as_of": "2026-10-13", "retrieved": "2026-10-15" }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v3", "value": 83.3, "unit": "%", "column": "Olmo 3.1 32B Think", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "as_of": "2026-10-13", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA", "value": 57.5, "unit": "%", "column": "Olmo 3.1 32B Think", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "as_of": "2026-10-13", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "olmo-3.1-32b-think", "value": 1287, "unit": "elo", "ci": 7, "votes": 8301, "rank": 299, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 7.1, "lmarena_elo": 1287 }, "popularity": { "ollama_pulls": "298.9K", "hf_downloads_30d": 19329, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/allenai/Olmo-3.1-32B-Think", "ollama": "https://ollama.com/library/olmo-3.1", "mlx_4bit": "https://huggingface.co/mlx-community/Olmo-3.1-32B-Think-4bit" }, "run": { "ollama": "ollama run olmo-3.1:32b-think", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Olmo-3.1-32B-Think-4bit --prompt 'hello'" }, "description": "Olmo is a series of Open language models designed to enable the science of language models. These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.", "notes": [], "gap": { "aa_points_behind_frontier": 50.5, "aa_pct_of_frontier": 12.3, "lmarena_elo_behind_frontier": 237, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "olmo-3.1-32b-instruct", "name": "OLMo 3.1 32B Instruct", "family": "OLMo 3", "vendor": "Ai2", "categories": [ "general", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2025-12-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 32.23, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 65536, "native_source": "HF config.json", "ollama_default": 65536, "ollama_source": "https://ollama.com/library/olmo-3.1" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "olmo-3.1:32b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 34.0, "tag": "olmo-3.1:32b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 64.0, "tag": "olmo-3.1:32b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "olmo-3.1:32b-instruct", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3.1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.13, "repo": "mlx-community/Olmo-3.1-32B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Olmo-3.1-32B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 41.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Olmo 3.1 32B Instruct", "value": 6.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Olmo 3.1 32B Instruct", "value": 53.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Olmo 3.1 32B Instruct", "value": 5.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Olmo 3.1 32B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Olmo 3.1 32B Instruct", "value": 21.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Olmo 3.1 32B Instruct", "value": 39.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Olmo 3.1 32B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-1-32b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "olmo-3.1-32b-instruct", "value": 1329, "unit": "elo", "ci": 6, "votes": 11817, "rank": 246, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 6.5, "lmarena_elo": 1329 }, "popularity": { "ollama_pulls": "298.9K", "hf_downloads_30d": 19401, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct", "ollama": "https://ollama.com/library/olmo-3.1", "mlx_4bit": "https://huggingface.co/mlx-community/Olmo-3.1-32B-Instruct-4bit" }, "run": { "ollama": "ollama run olmo-3.1:32b-instruct", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Olmo-3.1-32B-Instruct-4bit --prompt 'hello'" }, "description": "Olmo is a series of Open language models designed to enable the science of language models. These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.", "notes": [], "gap": { "aa_points_behind_frontier": 51.1, "aa_pct_of_frontier": 11.3, "lmarena_elo_behind_frontier": 195, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "olmo-3-7b-instruct", "name": "OLMo 3 7B Instruct", "family": "OLMo 3", "vendor": "Ai2", "categories": [ "general", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2025-11-19", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 7.3, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 65536, "native_source": "HF config.json", "ollama_default": 65536, "ollama_source": "https://ollama.com/library/olmo-3" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 4.5, "tag": "olmo-3:7b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 7.8, "tag": "olmo-3:7b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 15.0, "tag": "olmo-3:7b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 4.5, "tag": "olmo-3:7b-instruct", "source_name": "Ollama library", "url": "https://ollama.com/library/olmo-3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 4.11, "repo": "mlx-community/Olmo-3-7B-Instruct-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Olmo-3-7B-Instruct-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 4.5, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.2, "min_memory_gb_q8": 11.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Olmo 3 7B Instruct", "value": 5.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Olmo 3 7B Instruct", "value": 40.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Olmo 3 7B Instruct", "value": 5.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Olmo 3 7B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Olmo 3 7B Instruct", "value": 12.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Olmo 3 7B Instruct", "value": 32.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Olmo 3 7B Instruct", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-instruct", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Olmo 3 7B Think", "value": 5.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Olmo 3 7B Think", "value": 51.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Olmo 3 7B Think", "value": 6.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Olmo 3 7B Think", "value": 0.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Olmo 3 7B Think", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Olmo 3 7B Think", "value": 41.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Olmo 3 7B Think", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/olmo-3-7b-think", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 5.6, "lmarena_elo": null }, "popularity": { "ollama_pulls": "470.4K", "hf_downloads_30d": 492523, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/allenai/Olmo-3-7B-Instruct", "ollama": "https://ollama.com/library/olmo-3", "mlx_4bit": "https://huggingface.co/mlx-community/Olmo-3-7B-Instruct-4bit" }, "run": { "ollama": "ollama run olmo-3:7b-instruct", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Olmo-3-7B-Instruct-4bit --prompt 'hello'" }, "description": "Olmo is a series of Open language models designed to enable the science of language models. These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.", "notes": [], "gap": { "aa_points_behind_frontier": 52.0, "aa_pct_of_frontier": 9.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "smollm3-3b", "name": "SmolLM3 3B", "family": "SmolLM", "vendor": "Hugging Face", "categories": [ "edge", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2025-07-08", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 3.08, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 65536, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 1.92, "repo": "unsloth/SmolLM3-3B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/SmolLM3-3B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 3.28, "repo": "unsloth/SmolLM3-3B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/SmolLM3-3B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 1.73, "repo": "mlx-community/SmolLM3-3B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/SmolLM3-3B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 1.92, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 4.2, "min_memory_gb_q8": 5.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "LiveCodeBench", "variant": "Competitive programming LiveCodeBench v4", "value": 30.0, "unit": "%", "column": "SmoLLM3-3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B", "as_of": "2025-09-10", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 762049, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/SmolLM3-3B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/SmolLM3-3B-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "smollm2-1.7b", "name": "SmolLM2 1.7B", "family": "SmolLM", "vendor": "Hugging Face", "categories": [ "edge", "fully-open" ], "status": "legacy", "open_weights": true, "release_date": "2024-10-31", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1.71, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 8192, "native_source": "HF config.json", "ollama_default": 8192, "ollama_source": "https://ollama.com/library/smollm2" }, "inputs": null, "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 1.1, "tag": "smollm2:1.7b-instruct-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/smollm2/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 1.8, "tag": "smollm2:1.7b-instruct-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/smollm2/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 3.4, "tag": "smollm2:1.7b-instruct-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/smollm2/tags", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 1.1, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 3.3, "min_memory_gb_q8": 4.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro (MCF)", "value": 19.4, "unit": "%", "column": "SmolLM2-1.7B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct", "as_of": "2025-04-21", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "smollm2-1.7b-instruct", "value": 1115, "unit": "elo", "ci": 14, "votes": 2199, "rank": 397, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": null, "lmarena_elo": 1115 }, "popularity": { "ollama_pulls": "4.1M", "hf_downloads_30d": 233563, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct", "ollama": "https://ollama.com/library/smollm2", "mlx_4bit": null }, "run": { "ollama": "ollama run smollm2:1.7b-instruct", "ollama_cloud_only": false, "mlx": null }, "description": "SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": 409, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "minimax-m3", "name": "MiniMax M3", "family": "MiniMax", "vendor": "MiniMax", "categories": [ "coding", "agentic", "vision", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-06-02", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 427.04, "active_b": 23, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"~428B parameters and ~23B activated parameters\" (https://huggingface.co/MiniMaxAI/MiniMax-M3)" }, "license": { "id": "other", "name": "minimax-community", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 524288, "ollama_source": "https://ollama.com/library/minimax-m3" }, "inputs": null, "capabilities": [ "vision", "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 261.28, "repo": "bartowski/MiniMax-M3-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/MiniMax-M3-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 453.61, "repo": "unsloth/MiniMax-M3-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/MiniMax-M3-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 241.48, "repo": "mlx-community/MiniMax-M3-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/MiniMax-M3-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 261.28, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 302.5, "min_memory_gb_q8": 523.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiniMax-M3", "value": 29.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "MiniMax-M3", "value": 92.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "MiniMax-M3", "value": 39.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "MiniMax-M3", "value": 42.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "MiniMax-M3", "value": 65.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "MiniMax-M3", "value": 88.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "MiniMax-M3", "value": 47.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "MiniMax-M3", "value": 82.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "MiniMax-M3", "value": 83.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "MiniMax-M3", "value": 78.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m3", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LMArena Text Elo", "variant": "minimax-m3", "value": 1441, "unit": "elo", "ci": 4, "votes": 63973, "rank": 96, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 29.2, "lmarena_elo": 1441 }, "popularity": { "ollama_pulls": "523.7K", "hf_downloads_30d": 196809, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "ollama": "https://ollama.com/library/minimax-m3", "mlx_4bit": "https://huggingface.co/mlx-community/MiniMax-M3-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/MiniMax-M3-4bit --prompt 'hello'" }, "description": "MiniMax M3: Coding & Agentic Frontier. 1M context window. Native Multimodality.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows." ], "gap": { "aa_points_behind_frontier": 28.4, "aa_pct_of_frontier": 50.7, "lmarena_elo_behind_frontier": 83, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "minimax-m2.7", "name": "MiniMax M2.7", "family": "MiniMax", "vendor": "MiniMax", "categories": [ "coding", "agentic" ], "status": "superseded", "open_weights": true, "release_date": "2026-04-09", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 228.69, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "other", "url": "https://github.com/MiniMax-AI/MiniMax-M2.7/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 204800, "native_source": "HF config.json", "ollama_default": 204800, "ollama_source": "https://ollama.com/library/minimax-m2.7" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 138.34, "repo": "lmstudio-community/MiniMax-M2.7-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/lmstudio-community/MiniMax-M2.7-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 243.14, "repo": "unsloth/MiniMax-M2.7-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/MiniMax-M2.7-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 128.67, "repo": "mlx-community/MiniMax-M2.7-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/MiniMax-M2.7-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 138.34, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 161.1, "min_memory_gb_q8": 281.6, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiniMax-M2.7", "value": 22.8, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": true }, { "benchmark": "GPQA Diamond", "variant": "MiniMax-M2.7", "value": 87.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Humanity's Last Exam", "variant": "MiniMax-M2.7", "value": 29.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench Hard", "variant": "MiniMax-M2.7", "value": 39.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "Terminal-Bench 2.1", "variant": "MiniMax-M2.7", "value": 55.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "τ²-Bench Telecom", "variant": "MiniMax-M2.7", "value": 84.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "SciCode", "variant": "MiniMax-M2.7", "value": 50.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "IFBench", "variant": "MiniMax-M2.7", "value": 75.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "AA-LCR (long context)", "variant": "MiniMax-M2.7", "value": 78.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minimax-m2-7", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": true }, { "benchmark": "LMArena Text Elo", "variant": "minimax-m2.7", "value": 1415, "unit": "elo", "ci": 3, "votes": 84977, "rank": 140, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 22.8, "lmarena_elo": 1415 }, "popularity": { "ollama_pulls": "2.4M", "hf_downloads_30d": 657053, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "ollama": "https://ollama.com/library/minimax-m2.7", "mlx_4bit": "https://huggingface.co/mlx-community/MiniMax-M2.7-4bit" }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": "mlx_lm.generate --model mlx-community/MiniMax-M2.7-4bit --prompt 'hello'" }, "description": "MiniMax's M2-series model for coding, agentic workflows, and professional productivity.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-15; run locally via GGUF/MLX/vLLM if hardware allows.", "Artificial Analysis marks one or more listed variants as deprecated (older model); scores still shown on its site." ], "gap": { "aa_points_behind_frontier": 34.8, "aa_pct_of_frontier": 39.6, "lmarena_elo_behind_frontier": 109, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "hermes-4.3-36b", "name": "Hermes 4.3 36B", "family": "Hermes", "vendor": "Nous Research", "categories": [ "general", "finetune", "steerable" ], "status": "current", "open_weights": true, "release_date": "2025-11-17", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 36.15, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/NousResearch/Hermes-4.3-36B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 524288, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 21.76, "repo": "bartowski/NousResearch_Hermes-4.3-36B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_Hermes-4.3-36B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 38.42, "repo": "bartowski/NousResearch_Hermes-4.3-36B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_Hermes-4.3-36B-GGUF", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 21.76, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 27.0, "min_memory_gb_q8": 46.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 1641, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/NousResearch/Hermes-4.3-36B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "hermes-4-14b", "name": "Hermes 4 14B", "family": "Hermes", "vendor": "Nous Research", "categories": [ "general", "finetune", "steerable" ], "status": "current", "open_weights": true, "release_date": "2025-08-30", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 14, "active_b": null, "is_moe": false, "total_source": "parsed from model name (HF safetensors metadata incomplete)", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/NousResearch/Hermes-4-14B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 40960, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 9.0, "repo": "bartowski/NousResearch_Hermes-4-14B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_Hermes-4-14B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 15.7, "repo": "bartowski/NousResearch_Hermes-4-14B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_Hermes-4-14B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 8.31, "repo": "mlx-community/Hermes-4-14B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Hermes-4-14B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 9.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 12.3, "min_memory_gb_q8": 20.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 2991, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/NousResearch/Hermes-4-14B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/Hermes-4-14B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/Hermes-4-14B-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "hermes-4-70b", "name": "Hermes 4 70B (Llama 3.1)", "family": "Hermes", "vendor": "Nous Research", "categories": [ "general", "finetune", "steerable" ], "status": "current", "open_weights": true, "release_date": "2025-08-18", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 70.55, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "llama3", "name": null, "url": "https://huggingface.co/NousResearch/Hermes-4-70B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 42.52, "repo": "unsloth/Hermes-4-70B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Hermes-4-70B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 74.98, "repo": "unsloth/Hermes-4-70B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Hermes-4-70B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 39.69, "repo": "mlx-community/Hermes-4-70B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Hermes-4-70B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 42.52, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 50.9, "min_memory_gb_q8": 88.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 7.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 69.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 8.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 22.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 31.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Hermes 4 - Llama-3.1 70B (Reasoning)", "value": 9.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b-reasoning", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA Intelligence Index", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 6.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 49.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 3.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 21.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 29.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "Hermes 4 - Llama-3.1 70B (Non-reasoning)", "value": 4.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/hermes-4-llama-3-1-70b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 7.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 1498, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/NousResearch/Hermes-4-70B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/Hermes-4-70B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/Hermes-4-70B-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 49.7, "aa_pct_of_frontier": 13.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "hermes-3-8b", "name": "Hermes 3 8B (Llama 3.1)", "family": "Hermes", "vendor": "Nous Research", "categories": [ "general", "finetune", "steerable" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-07-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.03, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "llama3", "name": null, "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/hermes3" }, "inputs": null, "capabilities": [ "tools" ], "files": { "q4_k_m": { "value_gb": 4.9, "tag": "hermes3:8b-llama3.1-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/hermes3/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 8.5, "tag": "hermes3:8b-llama3.1-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/hermes3/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 16.0, "tag": "hermes3:8b-llama3.1-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/hermes3/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 4.52, "repo": "mlx-community/Hermes-3-Llama-3.1-8B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Hermes-3-Llama-3.1-8B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 4.9, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.6, "min_memory_gb_q8": 11.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.7M", "hf_downloads_30d": 508251, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B", "ollama": "https://ollama.com/library/hermes3", "mlx_4bit": "https://huggingface.co/mlx-community/Hermes-3-Llama-3.1-8B-4bit" }, "run": { "ollama": "ollama run hermes3:8b-llama3.1", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Hermes-3-Llama-3.1-8B-4bit --prompt 'hello'" }, "description": "Hermes 3 is the latest version of the flagship Hermes series of LLMs by Nous Research", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nouscoder-14b", "name": "NousCoder 14B", "family": "Hermes", "vendor": "Nous Research", "categories": [ "coding", "finetune" ], "status": "current", "open_weights": true, "release_date": "2025-12-23", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 14.77, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/NousResearch/NousCoder-14B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 81920, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 9.0, "repo": "bartowski/NousResearch_NousCoder-14B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_NousCoder-14B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 15.7, "repo": "bartowski/NousResearch_NousCoder-14B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/NousResearch_NousCoder-14B-GGUF", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 9.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 12.3, "min_memory_gb_q8": 20.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 508, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/NousResearch/NousCoder-14B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "laguna-xs-2.1", "name": "Laguna XS 2.1 33B-A3B", "family": "Laguna", "vendor": "Poolside", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-06-20", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 33.44, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "openmdw-1.1", "name": null, "url": "https://huggingface.co/poolside/Laguna-XS-2.1", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/laguna-xs-2.1" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 20.0, "tag": "laguna-xs-2.1:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 36.0, "tag": "laguna-xs-2.1:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 67.0, "tag": "laguna-xs-2.1:bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 20.0, "tag": "laguna-xs-2.1:nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 40.0, "tag": "laguna-xs-2.1:mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 20.0, "tag": "laguna-xs-2.1:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-xs-2.1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.82, "repo": "mlx-community/Laguna-XS-2.1-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Laguna-XS-2.1-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 20.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 25.0, "min_memory_gb_q8": 43.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "132.4K", "hf_downloads_30d": 395439, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/poolside/Laguna-XS-2.1", "ollama": "https://ollama.com/library/laguna-xs-2.1", "mlx_4bit": "https://huggingface.co/mlx-community/Laguna-XS-2.1-4bit" }, "run": { "ollama": "ollama run laguna-xs-2.1", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Laguna-XS-2.1-4bit --prompt 'hello'" }, "description": "Laguna XS 2.1 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine. ", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "laguna-s-2.1", "name": "Laguna S 2.1 118B-A8B", "family": "Laguna", "vendor": "Poolside", "categories": [ "coding", "agentic" ], "status": "current", "open_weights": true, "release_date": "2026-07-13", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 117.56, "active_b": 8.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "openmdw-1.1", "name": null, "url": "https://huggingface.co/poolside/Laguna-S-2.1", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/laguna-s-2.1" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 96.0, "tag": "laguna-s-2.1:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 128.0, "tag": "laguna-s-2.1:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 235.0, "tag": "laguna-s-2.1:f16", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 66.0, "tag": "laguna-s-2.1:nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 136.0, "tag": "laguna-s-2.1:mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 96.0, "tag": "laguna-s-2.1:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/laguna-s-2.1/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 96.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 112.4, "min_memory_gb_q8": 149.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "174.4K", "hf_downloads_30d": 328120, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/poolside/Laguna-S-2.1", "ollama": "https://ollama.com/library/laguna-s-2.1", "mlx_4bit": null }, "run": { "ollama": "ollama run laguna-s-2.1", "ollama_cloud_only": false, "mlx": null }, "description": "Our most capable model to date, designed for long-horizon work. 70.2% on Terminal-Bench 2.1 at 118B-A8B.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "north-mini-code-1.0", "name": "North Mini Code 1.0 30B-A3B", "family": "North", "vendor": "Cohere", "categories": [ "coding" ], "status": "current", "open_weights": true, "release_date": "2026-06-05", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 30.48, "active_b": 3.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 500000, "native_source": "HF config.json", "ollama_default": 499712, "ollama_source": "https://ollama.com/library/north-mini-code-1.0" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 19.0, "tag": "north-mini-code-1.0:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 32.0, "tag": "north-mini-code-1.0:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 61.0, "tag": "north-mini-code-1.0:bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15" }, "ollama_mlx": { "value_gb": 20.0, "tag": "north-mini-code-1.0:mlx-nvfp4", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15" }, "ollama_mxfp8": { "value_gb": 31.0, "tag": "north-mini-code-1.0:mlx-mxfp8", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 19.0, "tag": "north-mini-code-1.0:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/north-mini-code-1.0/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 18.5, "repo": "mlx-community/North-Mini-Code-1.0-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/North-Mini-Code-1.0-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 19.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 23.8, "min_memory_gb_q8": 38.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "North Mini Code", "value": 9.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "North Mini Code", "value": 75.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "North Mini Code", "value": 11.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "North Mini Code", "value": 31.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "North Mini Code", "value": 35.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "North Mini Code", "value": 37.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "North Mini Code", "value": 38.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "North Mini Code", "value": 57.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "North Mini Code", "value": 37.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/north-mini-code", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 9.9, "lmarena_elo": null }, "popularity": { "ollama_pulls": "71.6K", "hf_downloads_30d": 7233, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "ollama": "https://ollama.com/library/north-mini-code-1.0", "mlx_4bit": "https://huggingface.co/mlx-community/North-Mini-Code-1.0-4bit" }, "run": { "ollama": "ollama run north-mini-code-1.0", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/North-Mini-Code-1.0-4bit --prompt 'hello'" }, "description": "North Mini Code is Cohere's first model for developers — a 30B Mixture-of-Experts model with 3B active parameters, built for agentic software engineering.", "notes": [], "gap": { "aa_points_behind_frontier": 47.7, "aa_pct_of_frontier": 17.2, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "ornith-1.5-9b", "name": "Ornith 1.5 9B", "family": "Ornith", "vendor": "Ornith AI", "categories": [ "coding", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-08-19", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 9.0, "active_b": null, "is_moe": false, "total_source": "parsed from model name", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 262144, "ollama_source": "https://ollama.com/library/ornith-1.5" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "ollama_default": { "value_gb": 6.6, "tag": "ornith-1.5:9b", "source_name": "Ollama library", "url": "https://ollama.com/library/ornith-1.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 5.91, "repo": "bartowski/Ornith-1.5-9B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/Ornith-1.5-9B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.55, "repo": "bartowski/Ornith-1.5-9B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/Ornith-1.5-9B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 7.1, "repo": "mlx-community/Ornith-1.5-9B-OptiQ-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Ornith-1.5-9B-OptiQ-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.91, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.8, "min_memory_gb_q8": 13.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.1 (Terminus-2)", "value": 46.2, "unit": "%", "column": "Ornith-1.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 70.6, "unit": "%", "column": "Ornith-1.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 86.4, "unit": "%", "column": "Ornith-1.5-9B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "390.9K", "hf_downloads_30d": 5011351, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF", "ollama": "https://ollama.com/library/ornith-1.5", "mlx_4bit": "https://huggingface.co/mlx-community/Ornith-1.5-9B-OptiQ-4bit" }, "run": { "ollama": "ollama run ornith-1.5:9b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Ornith-1.5-9B-OptiQ-4bit --prompt 'hello'" }, "description": "Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.", "notes": [ "Total params parsed from name (repo is GGUF-only)." ], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "ornith-1.5-35b", "name": "Ornith 1.5 35B-A3B", "family": "Ornith", "vendor": "Ornith AI", "categories": [ "coding", "vision" ], "status": "current", "open_weights": true, "release_date": "2026-08-18", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 35.0, "active_b": 3, "is_moe": true, "total_source": "parsed from model name", "active_source": "\"~35B mixture-of-experts model with ~3B activated parameters\" (https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": 262144, "ollama_source": "https://ollama.com/library/ornith-1.5" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "ollama_default": { "value_gb": 23.0, "tag": "ornith-1.5:35b", "source_name": "Ollama library", "url": "https://ollama.com/library/ornith-1.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q4_k_m": { "value_gb": 21.86, "repo": "bartowski/Ornith-1.5-35B-A3B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/Ornith-1.5-35B-A3B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 37.81, "repo": "bartowski/Ornith-1.5-35B-A3B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/Ornith-1.5-35B-A3B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 23.55, "repo": "mlx-community/Ornith-1.5-35B-A3B-OptiQ-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/Ornith-1.5-35B-A3B-OptiQ-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 21.86, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 27.1, "min_memory_gb_q8": 45.5, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "Terminal-Bench", "variant": "Terminal-Bench 2.1 (Terminus-2)", "value": 67.8, "unit": "%", "column": "Ornith-1.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 79.0, "unit": "%", "column": "Ornith-1.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 89.2, "unit": "%", "column": "Ornith-1.5-35B-A3B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF", "as_of": "2026-08-24", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "390.9K", "hf_downloads_30d": 3709734, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF", "ollama": "https://ollama.com/library/ornith-1.5", "mlx_4bit": "https://huggingface.co/mlx-community/Ornith-1.5-35B-A3B-OptiQ-4bit" }, "run": { "ollama": "ollama run ornith-1.5:35b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/Ornith-1.5-35B-A3B-OptiQ-4bit --prompt 'hello'" }, "description": "Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "lfm2.5-8b-a1b", "name": "LFM2.5 8B-A1B", "family": "LFM2.5", "vendor": "Liquid AI", "categories": [ "edge", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2026-05-28", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.47, "active_b": 1.0, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "parsed from model name (AB)" }, "license": { "id": "other", "name": "lfm1.0", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 128000, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/lfm2.5" }, "inputs": null, "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 5.2, "tag": "lfm2.5:8b-a1b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.0, "tag": "lfm2.5:8b-a1b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 17.0, "tag": "lfm2.5:8b-a1b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 4.76, "repo": "mlx-community/LFM2.5-8B-A1B-MLX-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/LFM2.5-8B-A1B-MLX-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.2, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.0, "min_memory_gb_q8": 12.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "LFM2.5-8B-A1B", "value": 7.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "LFM2.5-8B-A1B", "value": 51.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "LFM2.5-8B-A1B", "value": 6.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "LFM2.5-8B-A1B", "value": 4.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "LFM2.5-8B-A1B", "value": 16.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "LFM2.5-8B-A1B", "value": 55.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "LFM2.5-8B-A1B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-8b-a1b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AIME 2025", "variant": "AIME25", "value": 42.53, "unit": "%", "column": "LFM2.5-8B-A1B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B", "as_of": "2026-08-24", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 7.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "209.6K", "hf_downloads_30d": 31712, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B", "ollama": "https://ollama.com/library/lfm2.5", "mlx_4bit": "https://huggingface.co/mlx-community/LFM2.5-8B-A1B-MLX-4bit" }, "run": { "ollama": "ollama run lfm2.5:8b-a1b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/LFM2.5-8B-A1B-MLX-4bit --prompt 'hello'" }, "description": "LFM2.5-8B-A1B, an edge model built for fast, reliable tool calling on consumer hardware.", "notes": [], "gap": { "aa_points_behind_frontier": 50.4, "aa_pct_of_frontier": 12.5, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "lfm2.5-24b-a2b", "name": "LFM2.5 24B-A2B", "family": "LFM2.5", "vendor": "Liquid AI", "categories": [ "general", "tool-calling", "agentic", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2026-10-12", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 23.97, "active_b": 2.45, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"Active parameters: 2.45B\" (https://huggingface.co/LiquidAI/LFM2.5-24B-A2B)" }, "license": { "id": "other", "name": "lfm1.0", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": 131072, "ollama_source": "https://ollama.com/library/lfm2.5" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 15.0, "tag": "lfm2.5:24b-a2b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 25.0, "tag": "lfm2.5:24b-a2b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 48.0, "tag": "lfm2.5:24b-a2b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5/tags", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 13.57, "repo": "LiquidAI/LFM2.5-24B-A2B-MLX-4bit", "source_name": "Hugging Face (Liquid AI official MLX 4-bit)", "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-MLX-4bit", "retrieved": "2026-10-16" } }, "memory": { "weights_gb_4bit": 15.0, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 19.2, "min_memory_gb_q8": 30.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 74.42, "unit": "%", "column": "LFM2.5-24B-A2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B", "as_of": "2026-10-13", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-Bench Verified (OpenCode)", "value": 51.0, "unit": "%", "column": "LFM2.5-24B-A2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B", "as_of": "2026-10-13", "retrieved": "2026-10-15" }, { "benchmark": "LiveClaw", "variant": "LiveClaw", "value": 48.09, "unit": "score", "column": "LFM2.5-24B-A2B", "source_type": "vendor-reported", "source_name": "Liquid AI blog", "url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "as_of": "2026-10-13", "retrieved": "2026-10-16" }, { "benchmark": "AutomationBench", "variant": "AutomationBench", "value": 27.69, "unit": "score", "column": "LFM2.5-24B-A2B", "source_type": "vendor-reported", "source_name": "Liquid AI blog", "url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "as_of": "2026-10-13", "retrieved": "2026-10-16" } ], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "213.8K", "hf_downloads_30d": 9356, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B", "ollama": "https://ollama.com/library/lfm2.5", "mlx_4bit": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-MLX-4bit" }, "run": { "ollama": "ollama run lfm2.5:24b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model LiquidAI/LFM2.5-24B-A2B-MLX-4bit --prompt 'hello'" }, "description": "Liquid AI's biggest LFM2.5 model: 24B in total but only about 2B work per word, built to run agents fast on your own machine.", "notes": [ "Reasoning only: it always thinks before it answers.", "Weak on knowledge questions alone. Liquid AI says pair it with web search or your files.", "Liquid AI measured 184 tokens/s on an M5 Max (llama.cpp, Q4_K_M, 4K input). Vendor number, not ours.", "Free licence up to $10M yearly revenue. Above that, commercial terms apply.", "The old MLX link pointed to an earlier Preview build. Now uses Liquid AI's own MLX 4-bit repo." ], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "lfm2.5-1.2b-thinking", "name": "LFM2.5 1.2B Thinking", "family": "LFM2.5", "vendor": "Liquid AI", "categories": [ "edge", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2026-01-20", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1.17, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "lfm1.0", "url": "LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 128000, "native_source": "HF config.json", "ollama_default": 128000, "ollama_source": "https://ollama.com/library/lfm2.5-thinking" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 0.731, "tag": "lfm2.5-thinking:1.2b-q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5-thinking/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 1.2, "tag": "lfm2.5-thinking:1.2b-q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5-thinking/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 2.3, "tag": "lfm2.5-thinking:1.2b-bf16", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5-thinking/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.731, "tag": "lfm2.5-thinking:1.2b", "source_name": "Ollama library", "url": "https://ollama.com/library/lfm2.5-thinking/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "mlx_4bit": { "value_gb": 0.66, "repo": "mlx-community/LFM2.5-1.2B-Thinking-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/LFM2.5-1.2B-Thinking-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.731, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 2.8, "min_memory_gb_q8": 3.4, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "LFM2.5-1.2B-Thinking", "value": 5.2, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "LFM2.5-1.2B-Thinking", "value": 33.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "LFM2.5-1.2B-Thinking", "value": 6.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "LFM2.5-1.2B-Thinking", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "LFM2.5-1.2B-Thinking", "value": 19.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "LFM2.5-1.2B-Thinking", "value": 41.8, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "LFM2.5-1.2B-Thinking", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/lfm2-5-1-2b-thinking", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 5.2, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.3M", "hf_downloads_30d": 8331, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking", "ollama": "https://ollama.com/library/lfm2.5-thinking", "mlx_4bit": "https://huggingface.co/mlx-community/LFM2.5-1.2B-Thinking-4bit" }, "run": { "ollama": "ollama run lfm2.5-thinking:1.2b", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/LFM2.5-1.2B-Thinking-4bit --prompt 'hello'" }, "description": "LFM2.5 is a new family of hybrid models designed for on-device deployment.", "notes": [], "gap": { "aa_points_behind_frontier": 52.4, "aa_pct_of_frontier": 9.0, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "minicpm-v-4.6", "name": "MiniCPM-V 4.6 (1B)", "family": "MiniCPM-V", "vendor": "OpenBMB", "categories": [ "vision", "edge" ], "status": "current", "open_weights": true, "release_date": "2026-04-13", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1.3, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/openbmb/MiniCPM-V-4.6", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": 262144, "ollama_source": "https://ollama.com/library/minicpm-v4.6" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 1.6, "tag": "minicpm-v4.6:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.6/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 1.9, "tag": "minicpm-v4.6:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.6/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 2.6, "tag": "minicpm-v4.6:f16", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.6/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 1.6, "tag": "minicpm-v4.6:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.6/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 1.6, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 3.8, "min_memory_gb_q8": 4.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiniCPM-V 4.6 1.3B", "value": 5.7, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": true, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "MiniCPM-V 4.6 1.3B", "value": 30.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "MiniCPM-V 4.6 1.3B", "value": 5.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench Hard", "variant": "MiniCPM-V 4.6 1.3B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "MiniCPM-V 4.6 1.3B", "value": 0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "τ²-Bench Telecom", "variant": "MiniCPM-V 4.6 1.3B", "value": 87.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "IFBench", "variant": "MiniCPM-V 4.6 1.3B", "value": 26.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "MiniCPM-V 4.6 1.3B", "value": 7.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "MiniCPM-V 4.6 1.3B", "value": 37.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm-v4-6-1-3b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 5.7, "lmarena_elo": null }, "popularity": { "ollama_pulls": "70.3K", "hf_downloads_30d": 207844, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/openbmb/MiniCPM-V-4.6", "ollama": "https://ollama.com/library/minicpm-v4.6", "mlx_4bit": null }, "run": { "ollama": "ollama run minicpm-v4.6", "ollama_cloud_only": false, "mlx": null }, "description": "A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone", "notes": [], "gap": { "aa_points_behind_frontier": 51.9, "aa_pct_of_frontier": 9.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "minicpm-v-4.5", "name": "MiniCPM-V 4.5 8B", "family": "MiniCPM-V", "vendor": "OpenBMB", "categories": [ "vision" ], "status": "legacy", "open_weights": true, "release_date": "2025-08-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.7, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/openbmb/MiniCPM-V-4_5", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 40960, "native_source": "HF config.json", "ollama_default": 40960, "ollama_source": "https://ollama.com/library/minicpm-v4.5" }, "inputs": "Text, Image", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 6.1, "tag": "minicpm-v4.5:q4_K_M", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.5/tags", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.8, "tag": "minicpm-v4.5:q8_0", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.5/tags", "retrieved": "2026-10-15" }, "bf16": { "value_gb": 17.0, "tag": "minicpm-v4.5:f16", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.5/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 6.1, "tag": "minicpm-v4.5:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/minicpm-v4.5/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 6.1, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 9.0, "min_memory_gb_q8": 13.3, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "55.1K", "hf_downloads_30d": 83708, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/openbmb/MiniCPM-V-4_5", "ollama": "https://ollama.com/library/minicpm-v4.5", "mlx_4bit": null }, "run": { "ollama": "ollama run minicpm-v4.5", "ollama_cloud_only": false, "mlx": null }, "description": "A GPT-4o Level MLLM for Single Image, Multi Image and High-FPS Video Understanding on Your Phone", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "minicpm5-2b", "name": "MiniCPM5 2B", "family": "MiniCPM", "vendor": "OpenBMB", "categories": [ "edge" ], "status": "current", "open_weights": true, "release_date": "2026-09-06", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 2.52, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 1.62, "repo": "bartowski/MiniCPM5-2B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/MiniCPM5-2B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 2.68, "repo": "bartowski/MiniCPM5-2B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/MiniCPM5-2B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 1.42, "repo": "mlx-community/MiniCPM5-2B-mlx-4Bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/MiniCPM5-2B-mlx-4Bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 1.62, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 3.9, "min_memory_gb_q8": 5.1, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiniCPM5-2B", "value": 12.5, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "MiniCPM5-2B", "value": 70.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "MiniCPM5-2B", "value": 8.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "MiniCPM5-2B", "value": 8.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "MiniCPM5-2B", "value": 26.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "MiniCPM5-2B", "value": 59.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/minicpm5-2b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "LiveCodeBench", "variant": "LiveCodeBench v6", "value": 69.1, "unit": "%", "column": "MiniCPM5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "as_of": "2026-09-29", "retrieved": "2026-10-15" }, { "benchmark": "AIME 2025", "variant": "AIME 2025", "value": 86.5, "unit": "%", "column": "MiniCPM5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "as_of": "2026-09-29", "retrieved": "2026-10-15" }, { "benchmark": "MMLU-Pro", "variant": "MMLU-Pro", "value": 70.8, "unit": "%", "column": "MiniCPM5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "as_of": "2026-09-29", "retrieved": "2026-10-15" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 46.4, "unit": "%", "column": "MiniCPM5-2B", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "as_of": "2026-09-29", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 12.5, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 1253288, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/openbmb/MiniCPM5-2B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/MiniCPM5-2B-mlx-4Bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/MiniCPM5-2B-mlx-4Bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 45.1, "aa_pct_of_frontier": 21.7, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mimo-v2.6-flash", "name": "MiMo-V2.6-Flash", "family": "MiMo", "vendor": "Xiaomi", "categories": [ "general", "coding", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-09-21", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 310.76, "active_b": 15, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"309B total / 15B activated parameters\" (https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": {}, "memory": { "weights_gb_4bit": 186.46, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 216.4, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiMo-V2.6-Flash", "value": 37.9, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "MiMo-V2.6-Flash", "value": 35.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "MiMo-V2.6-Flash", "value": 51.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "MiMo-V2.6-Flash", "value": 74.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "MMMU-Pro", "variant": "MiMo-V2.6-Flash", "value": 73.1, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-flash", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 4.0", "value": 28.8, "unit": "%", "column": "MiMo-V2.6 Flash", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "as_of": "2026-10-09", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "mimo-v2.6-flash", "value": 1451, "unit": "elo", "ci": 6, "votes": 10917, "rank": 77, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 37.9, "lmarena_elo": 1451 }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 84797, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 19.7, "aa_pct_of_frontier": 65.8, "lmarena_elo_behind_frontier": 73, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mimo-v2.6-pro", "name": "MiMo-V2.6-Pro", "family": "MiMo", "vendor": "Xiaomi", "categories": [ "general", "frontier-open" ], "status": "current", "open_weights": true, "release_date": "2026-09-21", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 1024.22, "active_b": 42, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"1.02T total / 42B activated parameters\" (https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL)" }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": {}, "memory": { "weights_gb_4bit": 614.53, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 708.7, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "MiMo-V2.6-Pro", "value": 46.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "MiMo-V2.6-Pro", "value": 49.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "MiMo-V2.6-Pro", "value": 60.9, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "MiMo-V2.6-Pro", "value": 86.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/mimo-v2-6-pro", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench", "variant": "Terminal Bench 4.0", "value": 34.9, "unit": "%", "column": "MiMo-V2.6 Pro", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "as_of": "2026-10-09", "retrieved": "2026-10-15" }, { "benchmark": "LMArena Text Elo", "variant": "mimo-v2.6-pro", "value": 1479, "unit": "elo", "ci": 9, "votes": 4075, "rank": 27, "preliminary": false, "source_type": "crowd-preference", "source_name": "LMArena", "url": "https://lmarena.ai/leaderboard/text", "as_of": "2026-10-15", "retrieved": "2026-10-15" } ], "summary": { "aa_index_best": 46.3, "lmarena_elo": 1479 }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 99548, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 11.3, "aa_pct_of_frontier": 80.4, "lmarena_elo_behind_frontier": 45, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "k2-horizon-7b", "name": "K2 Horizon 7B", "family": "K2 Horizon", "vendor": "IFM (MBZUAI)", "categories": [ "general", "fully-open" ], "status": "current", "open_weights": true, "release_date": "2026-09-01", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 9.0, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/IFM/K2-Horizon-7B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 524288, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 5.78, "repo": "bartowski/K2-Horizon-7B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/K2-Horizon-7B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 9.57, "repo": "bartowski/K2-Horizon-7B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/K2-Horizon-7B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 7.18, "repo": "mlx-community/K2-Horizon-7B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/K2-Horizon-7B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 5.78, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 8.6, "min_memory_gb_q8": 13.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "K2 Horizon 7B", "value": 20.6, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "K2 Horizon 7B", "value": 75.3, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "K2 Horizon 7B", "value": 18.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "K2 Horizon 7B", "value": 44.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "K2 Horizon 7B", "value": 27.5, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "K2 Horizon 7B", "value": 69.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-7b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 20.6, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 81808, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/IFM/K2-Horizon-7B", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/K2-Horizon-7B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/K2-Horizon-7B-4bit --prompt 'hello'" }, "description": null, "notes": [], "gap": { "aa_points_behind_frontier": 37.0, "aa_pct_of_frontier": 35.8, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "family": "K2 Horizon", "vendor": "IFM (MBZUAI)", "categories": [ "general" ], "status": "current", "open_weights": true, "release_date": "2026-09-03", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 36.0, "active_b": 4.0, "is_moe": true, "total_source": "parsed from model name", "active_source": "parsed from model name (AB)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B-GGUF", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": null, "native_source": null, "ollama_default": null, "ollama_source": null }, "inputs": null, "capabilities": [], "files": { "q4_k_m": { "value_gb": 22.7, "repo": "bartowski/K2-Horizon-MoVA-36B-A4B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/K2-Horizon-MoVA-36B-A4B-GGUF", "retrieved": "2026-10-15" }, "q8_0": { "value_gb": 39.83, "repo": "bartowski/K2-Horizon-MoVA-36B-A4B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/bartowski/K2-Horizon-MoVA-36B-A4B-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 26.4, "repo": "mlx-community/K2-Horizon-MoVA-36B-A4B-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/K2-Horizon-MoVA-36B-A4B-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 22.7, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 28.1, "min_memory_gb_q8": 47.8, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [ { "benchmark": "AA Intelligence Index", "variant": "K2 Horizon MoVA 36B A4B", "value": 25.3, "unit": "index", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": false, "aa_deprecated": false }, { "benchmark": "GPQA Diamond", "variant": "K2 Horizon MoVA 36B A4B", "value": 82.2, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Humanity's Last Exam", "variant": "K2 Horizon MoVA 36B A4B", "value": 23.4, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "Terminal-Bench 2.1", "variant": "K2 Horizon MoVA 36B A4B", "value": 56.6, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "SciCode", "variant": "K2 Horizon MoVA 36B A4B", "value": 40.0, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false }, { "benchmark": "AA-LCR (long context)", "variant": "K2 Horizon MoVA 36B A4B", "value": 71.7, "unit": "%", "source_type": "independent", "source_name": "Artificial Analysis", "url": "https://artificialanalysis.ai/models/k2-horizon-mova-36b-a4b", "as_of": "2026-10-15", "retrieved": "2026-10-15", "estimated": null, "aa_deprecated": false } ], "summary": { "aa_index_best": 25.3, "lmarena_elo": null }, "popularity": { "ollama_pulls": null, "hf_downloads_30d": 37182, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B-GGUF", "ollama": null, "mlx_4bit": "https://huggingface.co/mlx-community/K2-Horizon-MoVA-36B-A4B-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model mlx-community/K2-Horizon-MoVA-36B-A4B-4bit --prompt 'hello'" }, "description": null, "notes": [ "Total params parsed from name (repo is GGUF-only)." ], "gap": { "aa_points_behind_frontier": 32.3, "aa_pct_of_frontier": 43.9, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nomic-embed-text-v1.5", "name": "nomic-embed-text v1.5", "family": "Nomic Embed", "vendor": "Nomic AI", "categories": [ "embedding" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-02-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.14, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/nomic-ai/nomic-embed-text-v1.5", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 2048, "native_source": "HF config.json", "ollama_default": 2048, "ollama_source": "https://ollama.com/library/nomic-embed-text" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "ollama_default": { "value_gb": 0.274, "tag": "nomic-embed-text:v1.5", "source_name": "Ollama library", "url": "https://ollama.com/library/nomic-embed-text/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 0.274, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 2.3, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "90.2M", "hf_downloads_30d": 13239230, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nomic-ai/nomic-embed-text-v1.5", "ollama": "https://ollama.com/library/nomic-embed-text", "mlx_4bit": null }, "run": { "ollama": "ollama run nomic-embed-text:v1.5", "ollama_cloud_only": false, "mlx": null }, "description": "A high-performing open embedding model with a large token context window.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "nomic-embed-text-v2-moe", "name": "nomic-embed-text v2 MoE", "family": "Nomic Embed", "vendor": "Nomic AI", "categories": [ "embedding" ], "status": "current", "open_weights": true, "release_date": "2025-02-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.48, "active_b": null, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/nomic-ai/nomic-embed-text-v2-moe", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 2048, "native_source": "HF config.json", "ollama_default": 512, "ollama_source": "https://ollama.com/library/nomic-embed-text-v2-moe" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "ollama_default": { "value_gb": 0.958, "tag": "nomic-embed-text-v2-moe:latest", "source_name": "Ollama library", "url": "https://ollama.com/library/nomic-embed-text-v2-moe/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 0.958, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 3.1, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "1.1M", "hf_downloads_30d": 1139268, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/nomic-ai/nomic-embed-text-v2-moe", "ollama": "https://ollama.com/library/nomic-embed-text-v2-moe", "mlx_4bit": null }, "run": { "ollama": "ollama run nomic-embed-text-v2-moe", "ollama_cloud_only": false, "mlx": null }, "description": "nomic-embed-text-v2-moe is a multilingual MoE text embedding model that excels at multilingual retrieval.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mxbai-embed-large", "name": "mxbai-embed-large 335M", "family": "mxbai", "vendor": "Mixedbread", "categories": [ "embedding" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-03-07", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.34, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 512, "native_source": "HF config.json", "ollama_default": 512, "ollama_source": "https://ollama.com/library/mxbai-embed-large" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "bf16": { "value_gb": 0.67, "tag": "mxbai-embed-large:335m-v1-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/mxbai-embed-large/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 0.67, "tag": "mxbai-embed-large:335m", "source_name": "Ollama library", "url": "https://ollama.com/library/mxbai-embed-large/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" } }, "memory": { "weights_gb_4bit": 0.67, "weights_basis": "ollama_default", "weights_estimated": false, "min_memory_gb": 2.8, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "15.8M", "hf_downloads_30d": 1975794, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1", "ollama": "https://ollama.com/library/mxbai-embed-large", "mlx_4bit": null }, "run": { "ollama": "ollama run mxbai-embed-large:335m", "ollama_cloud_only": false, "mlx": null }, "description": "State-of-the-art large embedding model from mixedbread.ai", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "bge-m3", "name": "BGE-M3 567M", "family": "BGE", "vendor": "BAAI", "categories": [ "embedding" ], "status": "legacy-popular", "open_weights": true, "release_date": "2024-01-27", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 0.568, "active_b": null, "is_moe": false, "total_source": "Ollama library lists 567m", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/BAAI/bge-m3", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 8194, "native_source": "HF config.json", "ollama_default": 8192, "ollama_source": "https://ollama.com/library/bge-m3" }, "inputs": "Text", "capabilities": [ "embedding" ], "files": { "bf16": { "value_gb": 1.2, "tag": "bge-m3:567m-fp16", "source_name": "Ollama library", "url": "https://ollama.com/library/bge-m3/tags", "retrieved": "2026-10-15" }, "ollama_default": { "value_gb": 1.2, "tag": "bge-m3:567m", "source_name": "Ollama library", "url": "https://ollama.com/library/bge-m3/tags", "retrieved": "2026-10-15", "note": "default tag (quant varies by model)" }, "q8_0": { "value_gb": 0.63, "repo": "ggml-org/bge-m3-Q8_0-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/ggml-org/bge-m3-Q8_0-GGUF", "retrieved": "2026-10-15" }, "mlx_4bit": { "value_gb": 0.32, "repo": "mlx-community/bge-m3-mlx-4bit", "source_name": "Hugging Face (MLX community 4-bit)", "url": "https://huggingface.co/mlx-community/bge-m3-mlx-4bit", "retrieved": "2026-10-15" } }, "memory": { "weights_gb_4bit": 0.32, "weights_basis": "mlx_4bit", "weights_estimated": false, "min_memory_gb": 2.4, "min_memory_gb_q8": 2.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "benchmarks": [], "summary": { "aa_index_best": null, "lmarena_elo": null }, "popularity": { "ollama_pulls": "7.7M", "hf_downloads_30d": 32325522, "retrieved": "2026-10-15" }, "links": { "huggingface": "https://huggingface.co/BAAI/bge-m3", "ollama": "https://ollama.com/library/bge-m3", "mlx_4bit": "https://huggingface.co/mlx-community/bge-m3-mlx-4bit" }, "run": { "ollama": "ollama run bge-m3:567m", "ollama_cloud_only": false, "mlx": "mlx_lm.generate --model mlx-community/bge-m3-mlx-4bit --prompt 'hello'" }, "description": "BGE-M3 is a new model from BAAI distinguished for its versatility in Multi-Functionality, Multi-Linguality, and Multi-Granularity.", "notes": [], "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "astabrief-8b", "name": "AstaBrief 8B", "family": "AstaBrief", "vendor": "Ai2", "categories": [ "finetune" ], "status": "current", "open_weights": true, "release_date": "2026-10-02", "release_date_basis": "Ai2 blog post on Hugging Face (open-sourcing date; the repo was created private on 2026-02-09)", "params": { "total_b": 8.19, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata of the base model Qwen/Qwen3-8B (Ai2 says it started from Qwen3-8B)", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/allenai/AstaBrief_8B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 40960, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 5.03, "repo": "mradermacher/AstaBrief_8B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/mradermacher/AstaBrief_8B-GGUF", "retrieved": "2026-10-16" }, "q8_0": { "value_gb": 8.71, "repo": "mradermacher/AstaBrief_8B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/mradermacher/AstaBrief_8B-GGUF", "retrieved": "2026-10-16" }, "bf16": { "value_gb": 16.38, "repo": "allenai/AstaBrief_8B", "source_name": "Hugging Face repo file sizes (safetensors)", "url": "https://huggingface.co/allenai/AstaBrief_8B", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 2046, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/allenai/AstaBrief_8B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": "A small Ai2 model that turns a research question and paper excerpts into a report with citations.", "notes": [ "Works best with Ai2's own prompt format (see the model card).", "Ai2 says most training and tests were done in 2025, so its comparison models are not today's best.", "The GGUF files are from a community repo (mradermacher), not from Ai2.", "Ai2 also shared the training data." ], "memory": { "weights_gb_4bit": 5.03, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.8, "min_memory_gb_q8": 12.0, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "kolibri-1", "name": "Kolibri 1", "family": "Kolibri", "vendor": "Aleph Alpha", "categories": [ "general", "reasoning", "tool-calling" ], "status": "current", "open_weights": true, "release_date": "2026-10-03", "release_date_basis": "Hugging Face model card (Release Date)", "params": { "total_b": 78.1, "active_b": 3.46, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "Hugging Face model card" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 47.45, "repo": "Hob-forge/Kolibri-1-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/Hob-forge/Kolibri-1-GGUF", "retrieved": "2026-10-16" }, "q8_0": { "value_gb": 83.14, "repo": "Hob-forge/Kolibri-1-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/Hob-forge/Kolibri-1-GGUF", "retrieved": "2026-10-16" }, "bf16": { "value_gb": 156.21, "repo": "Aleph-Alpha/Kolibri-1-BF16", "source_name": "Hugging Face repo file sizes (BF16 safetensors)", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1-BF16", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 17864, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/Aleph-Alpha/Kolibri-1", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": "A German and English mixture-of-experts model from Aleph Alpha that can reason step by step and call tools.", "notes": [ "The 4-bit and 8-bit GGUF files are from a community repo (Hob-forge), not from Aleph Alpha.", "Aleph Alpha says it works up to 1M tokens of context but recommends at most 262,144.", "The model card lists about 78 GB in FP8 and a smallest server setup of 2× A100 80 GB or 1× H200." ], "memory": { "weights_gb_4bit": 47.45, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 56.6, "min_memory_gb_q8": 97.6, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "beam-1.0-501b-a23b", "name": "Beam 1.0 501B A23B", "family": "Beam", "vendor": "Reflection", "categories": [ "general", "coding", "agentic", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2026-10-07", "release_date_basis": "Hugging Face repo createdAt (announced 2026-10-05, weights released later)", "params": { "total_b": 501.27, "active_b": 23.4, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "Hugging Face model card" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": 1048576, "ollama_source": "https://ollama.com/library/beam" }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q8_0": { "value_gb": 532.79, "repo": "unsloth/Beam-1.0-501B-A23B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Beam-1.0-501B-A23B-GGUF", "retrieved": "2026-10-16" }, "bf16": { "value_gb": 1002.67, "repo": "unsloth/Beam-1.0-501B-A23B-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/unsloth/Beam-1.0-501B-A23B-GGUF", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": "466", "hf_downloads_30d": 440, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/reflection/Beam-1.0-501B-A23B", "ollama": "https://ollama.com/library/beam", "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": true, "mlx": null }, "description": "Reflection's first open-weight model, a large mixture-of-experts model for coding, reasoning and agent work.", "notes": [ "Ollama only offers this as a :cloud model (not local) as of 2026-10-16.", "No plain Q4_K_M file yet (Unsloth has UD-Q4_K_XL), so the memory need is estimated from the parameter count.", "Reflection also ships FP8 and NVFP4 versions on Hugging Face.", "Text only." ], "memory": { "weights_gb_4bit": 300.76, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 347.9, "min_memory_gb_q8": 614.7, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "mellum2.1-12b-a2.5b", "name": "Mellum2.1 12B-A2.5B Thinking", "family": "Mellum", "vendor": "JetBrains", "categories": [ "coding", "agentic", "tool-calling", "reasoning" ], "status": "current", "open_weights": true, "release_date": "2026-10-08", "release_date_basis": "JetBrains blog launch date (Hugging Face repo createdAt 2026-09-20 was a private upload)", "params": { "total_b": 12.15, "active_b": 2.5, "is_moe": true, "total_source": "Hugging Face safetensors metadata", "active_source": "\"2.5B active parameters\" (https://blog.jetbrains.com/ai/2026/10/mellum2-1-gets-to-work-a-fast-open-model-for-coding-agents/)" }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 131072, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": "Text", "capabilities": [ "tools", "thinking" ], "files": { "q4_k_m": { "value_gb": 8.07, "repo": "JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "retrieved": "2026-10-16", "note": "file Mellum2.1-12B-A2.5B-Thinking-Q4_K_M.gguf" }, "q8_0": { "value_gb": 12.93, "repo": "JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "retrieved": "2026-10-16", "note": "file Mellum2.1-12B-A2.5B-Thinking-Q8_0.gguf" }, "bf16": { "value_gb": 24.31, "repo": "JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "source_name": "Hugging Face GGUF repo", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF", "retrieved": "2026-10-16", "note": "file Mellum2.1-12B-A2.5B-Thinking-BF16.gguf" } }, "benchmarks": [ { "benchmark": "LiveCodeBench v6", "variant": "LiveCodeBench v6", "value": 82.0, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" }, { "benchmark": "SWE-bench Verified", "variant": "SWE-bench Verified", "value": 47.0, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" }, { "benchmark": "SWE-bench Pro", "variant": "SWE-bench Pro", "value": 28.0, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" }, { "benchmark": "Terminal-Bench 2.1", "variant": "Terminal-Bench 2.1", "value": 17.4, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" }, { "benchmark": "BFCL v4", "variant": "BFCL v4", "value": 62.3, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" }, { "benchmark": "GPQA Diamond", "variant": "GPQA Diamond", "value": 64.6, "unit": "%", "column": "Mellum2.1 Thinking", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "as_of": "2026-10-07", "retrieved": "2026-10-16" } ], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 2939, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": "Mellum2.1 is JetBrains' open coding model, 12B in total with 2.5B active, trained to explore a codebase, edit files and check its own changes.", "notes": [ "Thinking model for agent work: working in a repo, running commands, calling tools.", "Not on Ollama on 2026-10-16. JetBrains' own GGUF files (llama.cpp, LM Studio) went up 2026-10-07.", "All scores are JetBrains' own, from the model card." ], "memory": { "weights_gb_4bit": 8.07, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 11.3, "min_memory_gb_q8": 16.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "youtu-parsing-omni", "name": "Youtu-Parsing-Omni 5B", "family": "Youtu-Parsing", "vendor": "Tencent", "categories": [ "vision", "ocr", "audio", "video" ], "status": "current", "open_weights": true, "release_date": "2026-10-09", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 5.33, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "other", "name": "youtu-parsing", "url": "https://huggingface.co/tencent/Youtu-Parsing-Omni/blob/main/LICENSE", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 1048576, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": "Text, Image, Audio, Video", "capabilities": [ "vision" ], "files": { "bf16": { "value_gb": 10.67, "repo": "tencent/Youtu-Parsing-Omni", "source_name": "Hugging Face repo file sizes", "url": "https://huggingface.co/tencent/Youtu-Parsing-Omni/tree/main", "retrieved": "2026-10-16" } }, "benchmarks": [ { "benchmark": "OmniDocBench", "variant": "OmniDocBench v1.6 (Overall)", "value": 96.96, "unit": "score", "column": "Youtu-Parsing-Omni", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/tencent/Youtu-Parsing-Omni", "as_of": "2026-10-16", "retrieved": "2026-10-16" }, { "benchmark": "OmniParsingBench", "variant": "OmniParsingBench (Avg.)", "value": 75.08, "unit": "score", "column": "Youtu-Parsing-Omni", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/tencent/Youtu-Parsing-Omni", "as_of": "2026-10-16", "retrieved": "2026-10-16" } ], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 1552, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/tencent/Youtu-Parsing-Omni", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": false, "mlx": null }, "description": "Youtu-Parsing-Omni is a small Tencent model that turns documents, pictures, charts, audio and video into structured text.", "notes": [ "Official setup needs NVIDIA CUDA graphics cards (vLLM or Transformers); no official Mac build.", "License says it is not meant for use in the European Union.", "No 4-bit file from the maker; only the 10.7 GB BF16 weights." ], "memory": { "weights_gb_4bit": 3.2, "weights_basis": "params×0.6 (no published 4-bit file)", "weights_estimated": true, "min_memory_gb": 5.7, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "vllm-sr-d3-27b", "name": "vLLM Semantic Router d3 27B", "family": "Decision 3.0", "vendor": "vLLM Semantic Router", "categories": [ "vision", "finetune" ], "status": "current", "open_weights": true, "release_date": "2026-10-10", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 26.09, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/vllm-sr/d3", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "HF config.json", "ollama_default": null, "ollama_source": null }, "inputs": "Text, Image, Video", "capabilities": [ "vision" ], "files": { "q4_k_m": { "value_gb": 19.4, "repo": "ggml-org/vllm-d3-GGUF", "source_name": "Hugging Face repo file sizes", "url": "https://huggingface.co/ggml-org/vllm-d3-GGUF/tree/main", "retrieved": "2026-10-16" }, "q8_0": { "value_gb": 28.6, "repo": "ggml-org/vllm-d3-GGUF", "source_name": "Hugging Face repo file sizes", "url": "https://huggingface.co/ggml-org/vllm-d3-GGUF/tree/main", "retrieved": "2026-10-16" }, "bf16": { "value_gb": 52.18, "repo": "vllm-sr/d3", "source_name": "Hugging Face repo file sizes", "url": "https://huggingface.co/vllm-sr/d3/tree/main", "retrieved": "2026-10-16" } }, "benchmarks": [ { "benchmark": "Jev Decision Index", "variant": "Jev Decision Index 0.3.2 (text board)", "value": 66.3, "unit": "index", "column": "d3", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/vllm-sr/d3", "as_of": "2026-10-16", "retrieved": "2026-10-16" }, { "benchmark": "Perception Test", "variant": "Perception Test (validation, internal evaluation)", "value": 77.4, "unit": "%", "column": "d3", "source_type": "vendor-reported", "source_name": "Model card (Hugging Face)", "url": "https://huggingface.co/vllm-sr/d3", "as_of": "2026-10-16", "retrieved": "2026-10-16" } ], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 3565, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/vllm-sr/d3", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": false, "mlx": null }, "description": "d3 is a decision model: you give it an input and questions, and it picks an option, says yes or no, or gives a score, with a probability for each answer.", "notes": [ "Not a chat model: it writes no text. In llama.cpp it runs with `llama serve -hf ggml-org/vllm-d3-GGUF` and answers on the /v1/systemone API, not normal chat.", "Built on Qwen3.8-27B. Smaller versions exist: d3-flash (8.39B), d3-mini, d3-nano, d3-lite and d3-edge (0.59B).", "The 4-bit file (19.4 GB) needs a separate picture-reader file (mmproj) for images." ], "memory": { "weights_gb_4bit": 19.4, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 24.3, "min_memory_gb_q8": 34.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "userrepgen-4b", "name": "UserRepGen 4B", "family": "UserRepGen", "vendor": "Microsoft", "categories": [ "finetune" ], "status": "current", "open_weights": true, "release_date": "2026-10-14", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 4.54, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata", "active_source": null }, "license": { "id": "mit", "name": null, "url": "https://huggingface.co/microsoft/UserRepGen-4B", "gated": false, "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 9216, "native_source": "\"9,216 tokens: up to 8,192 prompt tokens and 1,024 completion tokens\" (https://huggingface.co/microsoft/UserRepGen-4B)", "ollama_default": null, "ollama_source": null }, "inputs": "Text", "capabilities": [], "files": { "q4_k_m": { "value_gb": 2.73, "repo": "bartowski/UserRepGen-4B-GGUF", "source_name": "Hugging Face GGUF repo (community)", "url": "https://huggingface.co/bartowski/UserRepGen-4B-GGUF", "retrieved": "2026-10-16" }, "q8_0": { "value_gb": 4.49, "repo": "bartowski/UserRepGen-4B-GGUF", "source_name": "Hugging Face GGUF repo (community)", "url": "https://huggingface.co/bartowski/UserRepGen-4B-GGUF", "retrieved": "2026-10-16" }, "bf16": { "value_gb": 9.1, "repo": "microsoft/UserRepGen-4B", "source_name": "Hugging Face repo (official safetensors)", "url": "https://huggingface.co/microsoft/UserRepGen-4B", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 589, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/microsoft/UserRepGen-4B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": "A small Microsoft model, tuned from Qwen3.5-4B, that reads a chat and writes a short card about what the person wanted and how they behaved.", "notes": [ "Not a chat model. Its cards feed a separate AI that plays the user when you test a chatbot.", "Microsoft warns the cards can identify people. They are not anonymous data.", "Microsoft tested it only with vLLM on NVIDIA GPUs. No official Mac or GGUF build; the GGUF sizes are a community build.", "Text only." ], "memory": { "weights_gb_4bit": 2.73, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 5.1, "min_memory_gb_q8": 7.2, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "apertus-1.5-8b", "name": "Apertus 1.5 8B", "family": "Apertus", "vendor": "Swiss AI Initiative", "categories": [ "general", "fully-open", "vision", "audio", "long-context" ], "status": "current", "open_weights": true, "release_date": "2026-07-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 8.9, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata (includes the image and audio tokenizers)", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B", "gated": "auto", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "\"The context window grows to 262,144 tokens\" (https://huggingface.co/blog/swiss-ai/apertus-v15-release-notes)", "ollama_default": null, "ollama_source": null }, "inputs": "Text, Image, Audio", "capabilities": [ "vision", "thinking", "tools" ], "files": { "q4_k_m": { "value_gb": 5.06, "repo": "Colby/apertus-v1.5-8b-text-Q4_K_M-GGUF", "source_name": "Hugging Face GGUF repo (community, text only)", "url": "https://huggingface.co/Colby/apertus-v1.5-8b-text-Q4_K_M-GGUF", "retrieved": "2026-10-16" }, "q8_0": { "value_gb": 8.57, "repo": "andreasmartin/apertus-v1.5-8b-text-Q8_0-GGUF", "source_name": "Hugging Face GGUF repo (community, verified by Swiss AI)", "url": "https://huggingface.co/andreasmartin/apertus-v1.5-8b-text-Q8_0-GGUF", "retrieved": "2026-10-16" }, "mlx_4bit": { "value_gb": 4.53, "repo": "tokimoa/apertus-v1.5-8b-mlx-4bit", "source_name": "Hugging Face (community MLX 4-bit)", "url": "https://huggingface.co/tokimoa/apertus-v1.5-8b-mlx-4bit", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 621652, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B", "ollama": null, "mlx_4bit": "https://huggingface.co/tokimoa/apertus-v1.5-8b-mlx-4bit" }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": "mlx_lm.generate --model tokimoa/apertus-v1.5-8b-mlx-4bit --prompt 'hello'" }, "description": "A fully open Swiss model: weights, training data and recipe are all public. It reads text, pictures and sound and answers in text.", "notes": [ "Sound input is experimental, per Swiss AI.", "Official support runs through Swiss AI's own vLLM and Transformers builds; upstream support is in progress.", "Hugging Face asks you to accept terms before download.", "Swiss AI published release notes and a technical report on 2026-10-15.", "GGUF builds are community text-only versions. Swiss AI lists Andreas Martin's 8B build as verified (https://apertus-ai.org/docs/guides/ollama/)." ], "memory": { "weights_gb_4bit": 5.06, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 7.8, "min_memory_gb_q8": 11.9, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } }, { "id": "apertus-1.5-70b", "name": "Apertus 1.5 70B", "family": "Apertus", "vendor": "Swiss AI Initiative", "categories": [ "general", "fully-open", "vision", "audio", "long-context" ], "status": "current", "open_weights": true, "release_date": "2026-07-24", "release_date_basis": "Hugging Face repo createdAt", "params": { "total_b": 72.01, "active_b": null, "is_moe": false, "total_source": "Hugging Face safetensors metadata (includes the image and audio tokenizers)", "active_source": null }, "license": { "id": "apache-2.0", "name": null, "url": "https://huggingface.co/swiss-ai/Apertus-v1.5-70B", "gated": "auto", "source": "Hugging Face model card metadata" }, "context_tokens": { "native": 262144, "native_source": "\"The context window grows to 262,144 tokens\" (https://huggingface.co/blog/swiss-ai/apertus-v15-release-notes)", "ollama_default": null, "ollama_source": null }, "inputs": "Text, Image, Audio", "capabilities": [ "vision", "thinking", "tools" ], "files": { "q4_k_m": { "value_gb": 43.72, "repo": "katya228/Apertus-v1.5-70B-text-GGUF", "source_name": "Hugging Face GGUF repo (community, text only, not verified by Swiss AI)", "url": "https://huggingface.co/katya228/Apertus-v1.5-70B-text-GGUF", "retrieved": "2026-10-16" } }, "benchmarks": [], "popularity": { "ollama_pulls": null, "hf_downloads_30d": 3623, "retrieved": "2026-10-16" }, "links": { "huggingface": "https://huggingface.co/swiss-ai/Apertus-v1.5-70B", "ollama": null, "mlx_4bit": null }, "run": { "ollama": null, "ollama_cloud_only": null, "mlx": null }, "description": "A fully open Swiss model: weights, training data and recipe are all public. It reads text, pictures and sound and answers in text.", "notes": [ "Sound input is experimental, per Swiss AI.", "Official support runs through Swiss AI's own vLLM and Transformers builds; upstream support is in progress.", "Hugging Face asks you to accept terms before download.", "Swiss AI published release notes and a technical report on 2026-10-15.", "The only 4-bit GGUF is a community text-only build that Swiss AI has not verified." ], "memory": { "weights_gb_4bit": 43.72, "weights_basis": "q4_k_m", "weights_estimated": false, "min_memory_gb": 52.3, "min_memory_gb_q8": null, "recommended_tier_gb": null, "formula": "min_memory_gb = weights_gb × 1.15 + 2 (≈8–16K context KV cache + runtime overhead); a tier fits when min_memory_gb ≤ 0.75 × tier_gb", "is_estimate": true }, "summary": { "aa_index_best": null, "lmarena_elo": null }, "gap": { "aa_points_behind_frontier": null, "aa_pct_of_frontier": null, "lmarena_elo_behind_frontier": null, "basis": "derived: frontier = Claude Opus 5.5 (Max, Default Fallback) AA index 57.6; LMArena #1 Elo 1524 (2026-10-15)" } } ] }