Mac Studio (M5 Ultra) for local AI.
With 512 GB it runs GLM-5.3-Flash (needs 233 GB). 102 models fit. Published speed: gpt-oss-120b at 108.6 tokens/sec.
Memory96 / 256 / 512 GB
Shared between CPU and graphics.
Bandwidth1200 GB/s
Higher means faster answers.
PriceFrom about $5,499
As of 2026-10-15. Check before buying.
By memory option.
| Memory | Usable | Models that fit | Best three |
|---|---|---|---|
| 96 GB | 72 GB | 83 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |
| 256 GB | 192 GB | 95 | Qwen3.8-Flash-Next (125B-A6B) (140 GB), Qwen3.8 27B (22.7 GB), Nemotron 3.5 Super VL 120B-A12B (105.5 GB) |
| 512 GB | 384 GB | 102 | GLM-5.3-Flash (233 GB), Qwen3.8-Flash-Next (125B-A6B) (140 GB), MiMo-V2.6-Flash (216.4 GB) |
Published speeds.
Measured by others, with sources. Not our tests.
| Model | Tokens/sec | Engine | Setup | Source |
|---|---|---|---|---|
| gpt-oss-120b | 108.6 | oMLX v0.7.0 · MXFP4 (MLX) | M5 Ultra 80c, 256GB | community-measured · 2026-10-15 |
| gpt-oss-120b | 110.1 | oMLX v0.7.0 · MXFP4 (MLX) | M5 Ultra 80c, 256GB | community-measured · 2026-10-15 |
Short answers.
- What is the best local LLM for the Mac Studio (M5 Ultra)?
- With 512 GB, GLM-5.3-Flash (needs 233 GB, AA 41.8). 102 of 124 models fit. With 96 GB: Qwen3.8 27B.
- How fast is the Mac Studio (M5 Ultra) for local AI?
- Published: gpt-oss-120b at 108.6 tokens/sec (oMLX v0.7.0, M5 Ultra 80c, 256GB); gpt-oss-120b at 110.1 tokens/sec (oMLX v0.7.0, M5 Ultra 80c, 256GB). These are other people's measurements, not ours.
- How much does the Mac Studio (M5 Ultra) cost?
- From about $5,499 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.
Specs from Apple – Mac Studio tech specs (2026-10-15). Prices are approximate and dated; 2026 prices are inflated by the memory-chip shortage.
# Mac Studio (M5 Ultra) for local AI URL: https://insidethemachine.ai/machines/mac-studio-m5-ultra Type: Mac desktop. Chip: M5 Ultra (64- or 80-core GPU). Memory: unified, options 96, 256, 512 GB. Bandwidth: 1200 GB/s. Spec source: https://www.apple.com/mac-studio/specs/ (2026-10-15). Price: from about $5,499 (2026-10-15, approximate). Status: current. ## Which models fit | memory GB | usable GB | models that fit | best 3 | |---|---|---|---| | 96 | 72 | 83 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) | | 256 | 192 | 95 | Qwen3.8-Flash-Next (125B-A6B) (140 GB), Qwen3.8 27B (22.7 GB), Nemotron 3.5 Super VL 120B-A12B (105.5 GB) | | 512 | 384 | 102 | GLM-5.3-Flash (233 GB), Qwen3.8-Flash-Next (125B-A6B) (140 GB), MiMo-V2.6-Flash (216.4 GB) | ## Published speeds (not our tests) - gpt-oss-120b: 108.6 tok/s decode, 2829 tok/s prompt (oMLX v0.7.0, MXFP4 (MLX), M5 Ultra 80c, 256GB, community-measured) https://omlx.ai/benchmarks/performance/bcoqxlcv (2026-10-15) - gpt-oss-120b: 110.1 tok/s decode, 1049 tok/s prompt (oMLX v0.7.0, MXFP4 (MLX), M5 Ultra 80c, 256GB, community-measured) https://omlx.ai/benchmarks/performance/bcoqxlcv (2026-10-15) Notes: 256GB/512GB require the 36-core CPU/80-core GPU chip. Exact 256/512GB configured prices not captured (gap). ## Questions **What is the best local LLM for the Mac Studio (M5 Ultra)?** With 512 GB, GLM-5.3-Flash (needs 233 GB, AA 41.8). 102 of 124 models fit. With 96 GB: Qwen3.8 27B. **How fast is the Mac Studio (M5 Ultra) for local AI?** Published: gpt-oss-120b at 108.6 tokens/sec (oMLX v0.7.0, M5 Ultra 80c, 256GB); gpt-oss-120b at 110.1 tokens/sec (oMLX v0.7.0, M5 Ultra 80c, 256GB). These are other people's measurements, not ours. **How much does the Mac Studio (M5 Ultra) cost?** From about $5,499 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.