Log inCreate wallet

The best LLMs for a 64 GB Mac.

39 of 73 open models run well in 64 GB. The GPU can use about 48 GB of it by default. Speed depends on the chip: from 273 GB/s on the M4 Pro to 800 GB/s on the M1 Ultra.

Smartest that runs wellQwen3.8 27B8-bit · 28.7 GB · 4–22 tok/sBiggest that runs wellNemotron 3.5 Bolt 74B A9B4-bit · 41.6 GB · 24–125 tok/sFor codeQwen3 Coder 30B A3B Instruct8-bit · 32.4 GB · 35–183 tok/s

Macs with 64 GB

9 chips from M1 to M5 come with 64 GB. Open one for its own list and speeds.

Runs well 47

Fits with room for context, at 10 or more tokens a second on the M1 Ultra.

ModelVersionMemoryM4 ProM1 Ultra
Qwen3.8 27B27B · Intelligence 33.78-bit28.7 GB4–813–22
Nemotron 3.5 Bolt 74B A9B74B · 9B active · Intelligence 21.44-bit41.6 GB24–4371–125
Qwen3.6 35B A3B35B · 3B active · Intelligence 18.28-bit37.2 GB38–68113–200
Muse Glimmer 30B29.6B · Intelligence 17.58-bit31.5 GB4–711–20
Gemma 4 31B30.7B · Intelligence 14.78-bit32.6 GB4–711–20
Nemotron 3.5 Lightning 30B A3B30B · 3B active · Intelligence 12.98-bit31.9 GB38–68113–200
Granite 4.2 30B30B · Intelligence 12.78-bit31.9 GB4–711–20
MiniCPM5 2B2.52B · Intelligence 12.58-bit2.7 GB45–81133–237
Qwen3.5 9B9B · Intelligence 11.28-bit9.6 GB13–2338–67
Granite 4.2 8B8B · Intelligence 11.18-bit8.5 GB14–2642–75
Granite 4.2 3B3B · Intelligence 9.18-bit3.2 GB38–68113–200
gpt-oss-20b21B · 3.6B active · Intelligence 9native 4.25-bit13.8 GB65–115189–337
Nemotron 3 Nano 30B A3B30B · 3.5B active · Intelligence 8.98-bit31.9 GB33–5997–173
Gemma 4 E4B8B · Intelligence 8.98-bit8.5 GB26–4675–133
Ministral 3 8B8.9B · Intelligence 5.58-bit9.5 GB13–2338–67
Nemotron 3 Nano 4B3.97B · Intelligence 4.88-bit4.2 GB29–5286–152
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B active8-bit32.4 GB35–62103–183
GLM-4.7 Flash30B · 3B active8-bit31.9 GB38–68113–200
Qwen3.6 27B27B8-bit28.7 GB4–813–22
Gemma 4 26B A4B25.2B · 3.8B active8-bit26.8 GB31–5590–160
Devstral Small 2 (24B)24B8-bit25.5 GB5–914–25
Magistral Small 1.2 (2509)24B8-bit25.5 GB5–914–25
Mistral Small 3.2 (24B)24B8-bit25.5 GB5–914–25
LFM2.5 24B A2B24B · 2B active8-bit25.5 GB59–104171–305
Phi-4 Reasoning Vision 15B15B8-bit15.9 GB10–1823–40
Phi-4 (14B)14B8-bit14.9 GB11–1924–43
Gemma 4 12B Unified12B8-bit12.7 GB10–1728–50
Qwen3 8B8.2B8-bit8.7 GB14–2541–74
Llama 3.1 8B Instruct8B8-bit8.5 GB14–2642–75
Gemma 4 E2B5.1B8-bit5.4 GB51–91150–267
Qwen3.5 4B4B8-bit4.3 GB29–5184–149
Phi-4 mini instruct3.8B8-bit4 GB31–5590–160
Llama 3.2 3B Instruct3.21B8-bit3.4 GB36–64106–188
SmolLM3 3B3B8-bit3.2 GB38–68113–200
LFM2.5 2.6B2.69B8-bit2.9 GB42–75124–221
Qwen3.5 2B2B8-bit2.1 GB59–104171–305
Llama 3.2 1B Instruct1.23B8-bit1.3 GB95–168277–492
Qwen3.5 0.8B0.8B8-bit0.9 GB137–243400–711
Qwen3 0.6B0.6B8-bit0.6 GB205–364600–1067
Qwen3.5 122B A10B122B · 10B active · Intelligence 15.62-bit lower quality38.1 GB40–70116–206
gpt-oss-120b117B · 5.1B active · Intelligence 11.62-bit lower quality36.6 GB77–137225–400
Mistral Small 4 (119B)119B · 6.5B active · Intelligence 11.32-bit lower quality37.2 GB61–109180–320
Qwen3 Coder Next80B · 3B active · Intelligence 9.23-bit lower quality35 GB95–168277–492
Llama 4 Scout 17B-16E Instruct109B · 17B active · Intelligence 6.52-bit lower quality34.1 GB23–4168–121
Ornith 1.5 122B122B · 10B active2-bit lower quality38.1 GB40–70116–206
Nemotron 3 Super 120B A12B120B · 12B active2-bit lower quality37.5 GB32–5795–168
GLM-4.5 Air106B · 12B active2-bit lower quality33.1 GB32–5795–168

Runs, but slowly 3

Fits, but under 10 tokens a second: fine for batch work, slow for chat.

ModelVersionMemoryM4 ProM1 Ultra
Llama 3.3 70B Instruct70B4-bit39.4 GB3–69–16
Mistral Medium 3.5 (128B)128B · Intelligence 14.22-bit lower quality40 GB3–59–16
Devstral 2 (123B)123B2-bit lower quality38.4 GB3–69–17

Too big 23

Needs more memory than 64 GB, even at 2 bits.

Questions

What is the best LLM for a Mac with 64 GB?
Of the models with a published Intelligence Index, Qwen3.8 27B scores highest among those that run well in 64 GB (8-bit). The biggest that runs well is Nemotron 3.5 Bolt 74B A9B. For code: Qwen3 Coder 30B A3B Instruct.
Is 64 GB enough for local AI on a Mac?
39 of the 73 open models here run well in it at 4 bits or better. macOS lets the GPU use about 48 GB of it by default; the model and its context have to fit in that.
Can a 64 GB Mac run gpt-oss-20b?
Yes, native 4.25-bit, at about 65–115 tokens a second on an M4 Pro and 189–337 on an M1 Ultra.

Other sizes: 8 GB · 16 GB · 18 GB · 24 GB · 32 GB · 36 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB

Run them with MLX Harness (free, open source) or Apple’s mlx-lm in Terminal: step by step.

How it’s worked out

Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.

Best local LLMs for a 64 GB Mac (M1 to M5): what runs, how fast