Mac Studio (M5 Max) for local AI.
With 128 GB it runs Qwen3.8 27B (needs 22.7 GB). 90 models fit. Published speed: gpt-oss-120b at 78.2 tokens/sec.
Memory36 / 48 / 64 / 128 GB
Shared between CPU and graphics.
Bandwidth614 GB/s
Higher means faster answers. 460 GB/s on 32-core GPU (36GB only); 614 GB/s on 40-core GPU (48/64/128GB).
PriceFrom about $2,499
As of 2026-10-15. Check before buying.
By memory option.
| Memory | Usable | Models that fit | Best three |
|---|---|---|---|
| 36 GB | 27 GB | 67 | Qwen3.8 27B (22.7 GB), Qwen3.5 27B (21.5 GB), Qwen3.6 27B (21.5 GB) |
| 48 GB | 36 GB | 76 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |
| 64 GB | 48 GB | 77 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |
| 128 GB | 96 GB | 90 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |
Published speeds.
Measured by others, with sources. Not our tests.
| Model | Tokens/sec | Engine | Setup | Source |
|---|---|---|---|---|
| gpt-oss-120b | 78.2 | oMLX v0.3.9 · MXFP4-Q8 (MLX) | M5 Max 40c, 128GB | community-measured · 2026-10-15 |
| gpt-oss-120b | 70.5 | oMLX v0.3.9 · MXFP4-Q8 (MLX) | M5 Max 40c, 128GB | community-measured · 2026-10-15 |
Short answers.
- What is the best local LLM for the Mac Studio (M5 Max)?
- With 128 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 90 of 124 models fit. With 36 GB: Qwen3.8 27B.
- How fast is the Mac Studio (M5 Max) for local AI?
- Published: gpt-oss-120b at 78.2 tokens/sec (oMLX v0.3.9, M5 Max 40c, 128GB); gpt-oss-120b at 70.5 tokens/sec (oMLX v0.3.9, M5 Max 40c, 128GB). These are other people's measurements, not ours.
- How much does the Mac Studio (M5 Max) cost?
- From about $2,499 (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 Max) for local AI URL: https://insidethemachine.ai/machines/mac-studio-m5-max Type: Mac desktop. Chip: M5 Max (32- or 40-core GPU). Memory: unified, options 36, 48, 64, 128 GB. Bandwidth: 614 GB/s (460 GB/s on 32-core GPU (36GB only); 614 GB/s on 40-core GPU (48/64/128GB).). Spec source: https://www.apple.com/mac-studio/specs/ (2026-10-15). Price: from about $2,499 (2026-10-15, approximate). Status: current. ## Which models fit | memory GB | usable GB | models that fit | best 3 | |---|---|---|---| | 36 | 27 | 67 | Qwen3.8 27B (22.7 GB), Qwen3.5 27B (21.5 GB), Qwen3.6 27B (21.5 GB) | | 48 | 36 | 76 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) | | 64 | 48 | 77 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) | | 128 | 96 | 90 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) | ## Published speeds (not our tests) - gpt-oss-120b: 78.2 tok/s decode, 727.4 tok/s prompt (oMLX v0.3.9, MXFP4-Q8 (MLX), M5 Max 40c, 128GB, community-measured) https://omlx.ai/benchmarks/dgadv0k9 (2026-10-15) - gpt-oss-120b: 70.5 tok/s decode, 1576 tok/s prompt (oMLX v0.3.9, MXFP4-Q8 (MLX), M5 Max 40c, 128GB, community-measured) https://omlx.ai/benchmarks/dgadv0k9 (2026-10-15) Notes: Ryan's planned machine: 128GB config. 128GB requires the 40-core GPU chip. ## Questions **What is the best local LLM for the Mac Studio (M5 Max)?** With 128 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 90 of 124 models fit. With 36 GB: Qwen3.8 27B. **How fast is the Mac Studio (M5 Max) for local AI?** Published: gpt-oss-120b at 78.2 tokens/sec (oMLX v0.3.9, M5 Max 40c, 128GB); gpt-oss-120b at 70.5 tokens/sec (oMLX v0.3.9, M5 Max 40c, 128GB). These are other people's measurements, not ours. **How much does the Mac Studio (M5 Max) cost?** From about $2,499 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.