Log inCreate wallet

What can an M6 with 32 GB run?

29 of 73 open models run well on it. The GPU can use about 24 GB of its 32 GB by default, and it reads memory at 170 GB/s, which sets how fast it writes. Found in: Mac mini (M6 or M5 Pro).

Smartest that runs wellQwen3.6 35B A3B4-bit · 19.7 GB · 45–80 tok/sBiggest that runs wellQwen3.6 35B A3B4-bit · 19.7 GB · 45–80 tok/sFor codeQwen3 Coder 30B A3B Instruct4-bit · 17.2 GB · 40–72 tok/s

Runs well 29

Fits with room for context, at 10 or more tokens a second.

ModelVersionMemoryTokens/s
Qwen3.6 35B A3B35B · 3B active · Intelligence 18.24-bit19.7 GB45–80
Nemotron 3.5 Lightning 30B A3B30B · 3B active · Intelligence 12.94-bit16.9 GB45–80
MiniCPM5 2B2.52B · Intelligence 12.58-bit2.7 GB28–50
Qwen3.5 9B9B · Intelligence 11.26-bit7.3 GB10–19
Granite 4.2 8B8B · Intelligence 11.16-bit6.5 GB12–21
Granite 4.2 3B3B · Intelligence 9.18-bit3.2 GB24–43
gpt-oss-20b21B · 3.6B active · Intelligence 9native 4.25-bit13.8 GB40–72
Nemotron 3 Nano 30B A3B30B · 3.5B active · Intelligence 8.94-bit16.9 GB38–68
Gemma 4 E4B8B · Intelligence 8.98-bit8.5 GB16–28
Ministral 3 8B8.9B · Intelligence 5.56-bit7.2 GB11–19
Nemotron 3 Nano 4B3.97B · Intelligence 4.88-bit4.2 GB18–32
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B active4-bit17.2 GB40–72
GLM-4.7 Flash30B · 3B active4-bit16.9 GB45–80
Gemma 4 26B A4B25.2B · 3.8B active6-bit20.5 GB25–44
LFM2.5 24B A2B24B · 2B active6-bit19.5 GB48–85
Phi-4 (14B)14B4-bit7.9 GB10–17
Gemma 4 12B Unified12B4-bit6.7 GB11–20
Qwen3 8B8.2B6-bit6.7 GB11–20
Llama 3.1 8B Instruct8B6-bit6.5 GB12–21
Gemma 4 E2B5.1B8-bit5.4 GB32–57
Qwen3.5 4B4B8-bit4.3 GB18–32
Phi-4 mini instruct3.8B8-bit4 GB19–34
Llama 3.2 3B Instruct3.21B8-bit3.4 GB23–40
SmolLM3 3B3B8-bit3.2 GB24–43
LFM2.5 2.6B2.69B8-bit2.9 GB26–47
Qwen3.5 2B2B8-bit2.1 GB36–65
Llama 3.2 1B Instruct1.23B8-bit1.3 GB59–105
Qwen3.5 0.8B0.8B8-bit0.9 GB85–151
Qwen3 0.6B0.6B8-bit0.6 GB128–227

Runs, but slowly 10

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

ModelVersionMemoryTokens/s
Qwen3.8 27B27B · Intelligence 33.76-bit21.9 GB3–6
Muse Glimmer 30B29.6B · Intelligence 17.54-bit16.7 GB5–8
Gemma 4 31B30.7B · Intelligence 14.74-bit17.3 GB4–8
Granite 4.2 30B30B · Intelligence 12.74-bit16.9 GB5–8
Qwen3.6 27B27B6-bit21.9 GB3–6
Devstral Small 2 (24B)24B6-bit19.5 GB4–7
Magistral Small 1.2 (2509)24B6-bit19.5 GB4–7
Mistral Small 3.2 (24B)24B6-bit19.5 GB4–7
Phi-4 Reasoning Vision 15B15B8-bit15.9 GB5–9
Llama 3.3 70B Instruct70B2-bit lower quality21.9 GB3–6

Only just fits 2

Needs almost all memory: raise the GPU memory limit and keep the context short.

ModelVersionMemoryTokens/s
Nemotron 3.5 Bolt 74B A9B74B · 9B active · Intelligence 21.42-bit lower quality23.1 GB27–49
Qwen3 Coder Next80B · 3B active · Intelligence 9.22-bit lower quality25 GB85–151

Too big 32

Needs more memory than 32 GB, even at 2 bits.

Other Macs

Every Mac with 32 GB, from any chip.

The M6 with more or less memory:

Neighbouring chips:

Head to head:

Compare it with any Mac or GPU · The M6 for local AI · The best Mac for local LLMs · Every Mac

Questions

What is the best LLM for an M6 with 32 GB?
Of the models with a published Intelligence Index, Qwen3.6 35B A3B scores highest among those that run well here: 4-bit, about 45–80 tokens a second.
Can an M6 with 32 GB run Llama 3.3 70B?
Yes, but slowly: 2-bit (lower quality), at about 3–6 tokens a second.
How fast is gpt-oss-20b on an M6?
About 40–72 tokens a second (native 4.25-bit), from the chip’s 170 GB/s memory bandwidth.

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.

What LLMs can an M6 with 32 GB run? Models and speed