Log inCreate wallet

What can an M5 Pro with 48 GB run?

33 of 73 open models run well on it. The GPU can use about 36 GB of its 48 GB by default, and it reads memory at 307 GB/s, which sets how fast it writes. Found in: MacBook Pro (14-inch, M5 Pro or M5 Max), MacBook Pro (16-inch, M5 Pro or M5 Max), Mac mini (M6 or M5 Pro).

Smartest that runs wellQwen3.6 35B A3B6-bit · 28.4 GB · 58–102 tok/sBiggest that runs wellQwen3.6 35B A3B6-bit · 28.4 GB · 58–102 tok/sFor codeQwen3 Coder 30B A3B Instruct8-bit · 32.4 GB · 39–70 tok/s

Runs well 36

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

ModelVersionMemoryTokens/s
Qwen3.6 35B A3B35B · 3B active · Intelligence 18.26-bit28.4 GB58–102
Nemotron 3.5 Lightning 30B A3B30B · 3B active · Intelligence 12.98-bit31.9 GB43–77
MiniCPM5 2B2.52B · Intelligence 12.58-bit2.7 GB51–91
Qwen3.5 9B9B · Intelligence 11.28-bit9.6 GB14–26
Granite 4.2 8B8B · Intelligence 11.18-bit8.5 GB16–29
Granite 4.2 3B3B · Intelligence 9.18-bit3.2 GB43–77
gpt-oss-20b21B · 3.6B active · Intelligence 9native 4.25-bit13.8 GB73–129
Nemotron 3 Nano 30B A3B30B · 3.5B active · Intelligence 8.98-bit31.9 GB37–66
Gemma 4 E4B8B · Intelligence 8.98-bit8.5 GB29–51
Ministral 3 8B8.9B · Intelligence 5.58-bit9.5 GB15–26
Nemotron 3 Nano 4B3.97B · Intelligence 4.88-bit4.2 GB33–58
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B active8-bit32.4 GB39–70
GLM-4.7 Flash30B · 3B active8-bit31.9 GB43–77
Gemma 4 26B A4B25.2B · 3.8B active8-bit26.8 GB35–61
Devstral Small 2 (24B)24B4-bit13.5 GB10–18
Magistral Small 1.2 (2509)24B4-bit13.5 GB10–18
Mistral Small 3.2 (24B)24B4-bit13.5 GB10–18
LFM2.5 24B A2B24B · 2B active8-bit25.5 GB66–117
Phi-4 Reasoning Vision 15B15B6-bit12.2 GB11–20
Phi-4 (14B)14B6-bit11.4 GB12–22
Gemma 4 12B Unified12B8-bit12.7 GB11–19
Qwen3 8B8.2B8-bit8.7 GB16–28
Llama 3.1 8B Instruct8B8-bit8.5 GB16–29
Gemma 4 E2B5.1B8-bit5.4 GB58–102
Qwen3.5 4B4B8-bit4.3 GB32–57
Phi-4 mini instruct3.8B8-bit4 GB35–61
Llama 3.2 3B Instruct3.21B8-bit3.4 GB41–72
SmolLM3 3B3B8-bit3.2 GB43–77
LFM2.5 2.6B2.69B8-bit2.9 GB48–85
Qwen3.5 2B2B8-bit2.1 GB66–117
Llama 3.2 1B Instruct1.23B8-bit1.3 GB106–189
Qwen3.5 0.8B0.8B8-bit0.9 GB154–273
Qwen3 0.6B0.6B8-bit0.6 GB230–409
Nemotron 3.5 Bolt 74B A9B74B · 9B active · Intelligence 21.43-bit lower quality32.4 GB35–63
Qwen3 Coder Next80B · 3B active · Intelligence 9.22-bit lower quality25 GB154–273
GLM-4.5 Air106B · 12B active2-bit lower quality33.1 GB36–65

Runs, but slowly 6

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

ModelVersionMemoryTokens/s
Qwen3.8 27B27B · Intelligence 33.78-bit28.7 GB5–9
Muse Glimmer 30B29.6B · Intelligence 17.58-bit31.5 GB4–8
Gemma 4 31B30.7B · Intelligence 14.78-bit32.6 GB4–8
Granite 4.2 30B30B · Intelligence 12.78-bit31.9 GB4–8
Qwen3.6 27B27B8-bit28.7 GB5–9
Llama 3.3 70B Instruct70B3-bit lower quality30.6 GB5–8

Only just fits 8

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

ModelVersionMemoryTokens/s
Qwen3.5 122B A10B122B · 10B active · Intelligence 15.62-bit lower quality38.1 GB45–79
Mistral Medium 3.5 (128B)128B · Intelligence 14.22-bit lower quality40 GB3–6
gpt-oss-120b117B · 5.1B active · Intelligence 11.62-bit lower quality36.6 GB86–154
Mistral Small 4 (119B)119B · 6.5B active · Intelligence 11.32-bit lower quality37.2 GB69–123
Llama 4 Scout 17B-16E Instruct109B · 17B active · Intelligence 6.52-bit lower quality34.1 GB26–46
Devstral 2 (123B)123B2-bit lower quality38.4 GB4–6
Ornith 1.5 122B122B · 10B active2-bit lower quality38.1 GB45–79
Nemotron 3 Super 120B A12B120B · 12B active2-bit lower quality37.5 GB36–65

Too big 23

Needs more memory than 48 GB, even at 2 bits.

Other Macs

Every Mac with 48 GB, from any chip.

The M5 Pro with more or less memory:

Neighbouring chips:

Head to head:

Compare it with any Mac or GPU · The M5 Pro for local AI · The best Mac for local LLMs · Every Mac

Questions

What is the best LLM for an M5 Pro with 48 GB?
Of the models with a published Intelligence Index, Qwen3.6 35B A3B scores highest among those that run well here: 6-bit, about 58–102 tokens a second.
Can an M5 Pro with 48 GB run Llama 3.3 70B?
Yes, but slowly: 3-bit (lower quality), at about 5–8 tokens a second.
How fast is gpt-oss-20b on an M5 Pro?
About 73–129 tokens a second (native 4.25-bit), from the chip’s 307 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 M5 Pro with 48 GB run? Models and speed