Log inCreate wallet

What can an M3 Ultra with 256 GB run?

53 of 73 open models run well on it. The GPU can use about 192 GB of its 256 GB by default, and it reads memory at 819 GB/s, which sets how fast it writes. Found in: Mac Studio (2025).

Smartest that runs wellQwen3.8 Flash Next6-bit · 146 GB · 75–134 tok/sBiggest that runs wellMiMo V2.6 Flashnative 4.48-bit · 173 GB · 44–78 tok/sFor codeQwen3 Coder Next8-bit · 85 GB · 115–205 tok/s

Runs well 60

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

ModelVersionMemoryTokens/s
Qwen3.8 Flash Next180B · 6B active · Intelligence 39.86-bit146 GB75–134
MiMo V2.6 Flash309B · 15B active · Intelligence 37.9native 4.48-bit173 GB44–78
Qwen3.8 27B27B · Intelligence 33.78-bit28.7 GB13–23
Inkling Small276B · 12B active · Intelligence 25.74-bit155 GB54–96
Hy3295B · 21B active · Intelligence 25.34-bit166 GB31–56
Nemotron 3.5 Bolt 74B A9B74B · 9B active · Intelligence 21.48-bit78.6 GB38–68
Qwen3.6 35B A3B35B · 3B active · Intelligence 18.28-bit37.2 GB115–205
Muse Glimmer 30B29.6B · Intelligence 17.58-bit31.5 GB12–21
Qwen3.5 122B A10B122B · 10B active · Intelligence 15.68-bit130 GB35–62
Gemma 4 31B30.7B · Intelligence 14.78-bit32.6 GB11–20
Nemotron 3.5 Lightning 30B A3B30B · 3B active · Intelligence 12.98-bit31.9 GB115–205
Granite 4.2 30B30B · Intelligence 12.78-bit31.9 GB12–21
MiniCPM5 2B2.52B · Intelligence 12.58-bit2.7 GB137–243
gpt-oss-120b117B · 5.1B active · Intelligence 11.6native 4.25-bit65.3 GB137–243
Mistral Small 4 (119B)119B · 6.5B active · Intelligence 11.38-bit126 GB53–95
Qwen3.5 9B9B · Intelligence 11.28-bit9.6 GB38–68
Granite 4.2 8B8B · Intelligence 11.18-bit8.5 GB43–77
Qwen3 Coder Next80B · 3B active · Intelligence 9.28-bit85 GB115–205
Granite 4.2 3B3B · Intelligence 9.18-bit3.2 GB115–205
gpt-oss-20b21B · 3.6B active · Intelligence 9native 4.25-bit13.8 GB194–345
Nemotron 3 Nano 30B A3B30B · 3.5B active · Intelligence 8.98-bit31.9 GB100–177
Gemma 4 E4B8B · Intelligence 8.98-bit8.5 GB77–137
Llama 4 Scout 17B-16E Instruct109B · 17B active · Intelligence 6.58-bit116 GB20–36
Ministral 3 8B8.9B · Intelligence 5.58-bit9.5 GB39–69
Nemotron 3 Nano 4B3.97B · Intelligence 4.88-bit4.2 GB88–156
DeepSeek V4 Flash 0731284B · 13B active4-bit160 GB50–90
MiniMax M2.7230B · 10B active4-bit129 GB66–117
Ornith 1.5 122B122B · 10B active8-bit130 GB35–62
Nemotron 3 Super 120B A12B120B · 12B active8-bit128 GB29–51
GLM-4.5 Air106B · 12B active8-bit113 GB29–51
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B active8-bit32.4 GB105–187
GLM-4.7 Flash30B · 3B active8-bit31.9 GB115–205
Qwen3.6 27B27B8-bit28.7 GB13–23
Gemma 4 26B A4B25.2B · 3.8B active8-bit26.8 GB92–164
Devstral Small 2 (24B)24B8-bit25.5 GB14–26
Magistral Small 1.2 (2509)24B8-bit25.5 GB14–26
Mistral Small 3.2 (24B)24B8-bit25.5 GB14–26
LFM2.5 24B A2B24B · 2B active8-bit25.5 GB176–312
Phi-4 Reasoning Vision 15B15B8-bit15.9 GB23–41
Phi-4 (14B)14B8-bit14.9 GB25–44
Gemma 4 12B Unified12B8-bit12.7 GB29–52
Qwen3 8B8.2B8-bit8.7 GB42–75
Llama 3.1 8B Instruct8B8-bit8.5 GB43–77
Gemma 4 E2B5.1B8-bit5.4 GB154–273
Qwen3.5 4B4B8-bit4.3 GB86–152
Phi-4 mini instruct3.8B8-bit4 GB92–164
Llama 3.2 3B Instruct3.21B8-bit3.4 GB108–193
SmolLM3 3B3B8-bit3.2 GB115–205
LFM2.5 2.6B2.69B8-bit2.9 GB127–226
Qwen3.5 2B2B8-bit2.1 GB176–312
Llama 3.2 1B Instruct1.23B8-bit1.3 GB284–504
Qwen3.5 0.8B0.8B8-bit0.9 GB410–728
Qwen3 0.6B0.6B8-bit0.6 GB614–1092
DeepSeek V4.1 Flash 1017552B · 16B active · Intelligence 45.12-bit lower quality173 GB74–131
GLM-5.3 Flash320B · 18B active · Intelligence 41.83-bit lower quality140 GB47–83
Beam 1.0501B · 23B active · Intelligence 30.32-bit lower quality157 GB51–91
MiniMax M3428B · 23B active · Intelligence 29.22-bit lower quality134 GB51–91
Nemotron 3 Ultra 550B A55B550B · 55B active · Intelligence 22.92-bit lower quality172 GB21–38
Qwen3.5 397B A17B397B · 17B active · Intelligence 18.43-bit lower quality174 GB50–89
Llama 4 Maverick 17B-128E Instruct400B · 17B active · Intelligence 9.33-bit lower quality175 GB50–89

Runs, but slowly 3

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

ModelVersionMemoryTokens/s
Mistral Medium 3.5 (128B)128B · Intelligence 14.28-bit136 GB3–5
Devstral 2 (123B)123B8-bit131 GB3–5
Llama 3.3 70B Instruct70B8-bit74.4 GB5–9

Only just fits 1

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

ModelVersionMemoryTokens/s
Mistral Large 3 (675B)675B · 41B active2-bit lower quality211 GB29–51

Too big 9

Needs more memory than 256 GB, even at 2 bits.

Other Macs

Every Mac with 256 GB, from any chip.

The M3 Ultra with more or less memory:

Neighbouring chips:

Head to head:

Against PCs and graphics cards:

Compare it with any Mac or GPU · The M3 Ultra for local AI · The best Mac for local LLMs · Every Mac

Questions

What is the best LLM for an M3 Ultra with 256 GB?
Of the models with a published Intelligence Index, Qwen3.8 Flash Next scores highest among those that run well here: 6-bit, about 75–134 tokens a second. The biggest that runs well is MiMo V2.6 Flash.
Can an M3 Ultra with 256 GB run Llama 3.3 70B?
Yes, but slowly: 8-bit, at about 5–9 tokens a second.
How fast is gpt-oss-20b on an M3 Ultra?
About 194–345 tokens a second (native 4.25-bit), from the chip’s 819 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 M3 Ultra with 256 GB run? Models and speed