Log inCreate wallet

What can an M5 Ultra with 256 GB run?

54 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 1,200 GB/s, which sets how fast it writes. Found in: Mac Studio (M5 Max or M5 Ultra).

Smartest that runs wellQwen3.8 Flash Next6-bit · 146 GB · 110–196 tok/sBiggest that runs wellMiMo V2.6 Flashnative 4.48-bit · 173 GB · 64–114 tok/sFor codeQwen3 Coder Next8-bit · 85 GB · 169–300 tok/s

Runs well 61

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

ModelVersionMemoryTokens/s
Qwen3.8 Flash Next180B · 6B active · Intelligence 39.86-bit146 GB110–196
MiMo V2.6 Flash309B · 15B active · Intelligence 37.9native 4.48-bit173 GB64–114
Qwen3.8 27B27B · Intelligence 33.78-bit28.7 GB19–33
Inkling Small276B · 12B active · Intelligence 25.74-bit155 GB79–141
Hy3295B · 21B active · Intelligence 25.34-bit166 GB46–81
Nemotron 3.5 Bolt 74B A9B74B · 9B active · Intelligence 21.48-bit78.6 GB56–100
Qwen3.6 35B A3B35B · 3B active · Intelligence 18.28-bit37.2 GB169–300
Muse Glimmer 30B29.6B · Intelligence 17.58-bit31.5 GB17–30
Qwen3.5 122B A10B122B · 10B active · Intelligence 15.68-bit130 GB51–91
Gemma 4 31B30.7B · Intelligence 14.78-bit32.6 GB17–29
Nemotron 3.5 Lightning 30B A3B30B · 3B active · Intelligence 12.98-bit31.9 GB169–300
Granite 4.2 30B30B · Intelligence 12.78-bit31.9 GB17–30
MiniCPM5 2B2.52B · Intelligence 12.58-bit2.7 GB200–356
gpt-oss-120b117B · 5.1B active · Intelligence 11.6native 4.25-bit65.3 GB200–356
Mistral Small 4 (119B)119B · 6.5B active · Intelligence 11.38-bit126 GB78–139
Qwen3.5 9B9B · Intelligence 11.28-bit9.6 GB56–100
Granite 4.2 8B8B · Intelligence 11.18-bit8.5 GB64–113
Qwen3 Coder Next80B · 3B active · Intelligence 9.28-bit85 GB169–300
Granite 4.2 3B3B · Intelligence 9.18-bit3.2 GB169–300
gpt-oss-20b21B · 3.6B active · Intelligence 9native 4.25-bit13.8 GB284–505
Nemotron 3 Nano 30B A3B30B · 3.5B active · Intelligence 8.98-bit31.9 GB146–259
Gemma 4 E4B8B · Intelligence 8.98-bit8.5 GB113–200
Llama 4 Scout 17B-16E Instruct109B · 17B active · Intelligence 6.58-bit116 GB30–53
Ministral 3 8B8.9B · Intelligence 5.58-bit9.5 GB57–101
Nemotron 3 Nano 4B3.97B · Intelligence 4.88-bit4.2 GB129–229
DeepSeek V4 Flash 0731284B · 13B active4-bit160 GB74–132
MiniMax M2.7230B · 10B active4-bit129 GB96–171
Ornith 1.5 122B122B · 10B active8-bit130 GB51–91
Nemotron 3 Super 120B A12B120B · 12B active8-bit128 GB42–75
GLM-4.5 Air106B · 12B active8-bit113 GB42–75
Llama 3.3 70B Instruct70B4-bit39.4 GB14–24
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B active8-bit32.4 GB154–274
GLM-4.7 Flash30B · 3B active8-bit31.9 GB169–300
Qwen3.6 27B27B8-bit28.7 GB19–33
Gemma 4 26B A4B25.2B · 3.8B active8-bit26.8 GB135–240
Devstral Small 2 (24B)24B8-bit25.5 GB21–38
Magistral Small 1.2 (2509)24B8-bit25.5 GB21–38
Mistral Small 3.2 (24B)24B8-bit25.5 GB21–38
LFM2.5 24B A2B24B · 2B active8-bit25.5 GB257–457
Phi-4 Reasoning Vision 15B15B8-bit15.9 GB34–60
Phi-4 (14B)14B8-bit14.9 GB36–64
Gemma 4 12B Unified12B8-bit12.7 GB43–76
Qwen3 8B8.2B8-bit8.7 GB62–110
Llama 3.1 8B Instruct8B8-bit8.5 GB64–113
Gemma 4 E2B5.1B8-bit5.4 GB225–400
Qwen3.5 4B4B8-bit4.3 GB126–223
Phi-4 mini instruct3.8B8-bit4 GB135–240
Llama 3.2 3B Instruct3.21B8-bit3.4 GB159–282
SmolLM3 3B3B8-bit3.2 GB169–300
LFM2.5 2.6B2.69B8-bit2.9 GB186–331
Qwen3.5 2B2B8-bit2.1 GB257–457
Llama 3.2 1B Instruct1.23B8-bit1.3 GB415–738
Qwen3.5 0.8B0.8B8-bit0.9 GB600–1067
Qwen3 0.6B0.6B8-bit0.6 GB900–1600
DeepSeek V4.1 Flash 1017552B · 16B active · Intelligence 45.12-bit lower quality173 GB108–192
GLM-5.3 Flash320B · 18B active · Intelligence 41.83-bit lower quality140 GB68–122
Beam 1.0501B · 23B active · Intelligence 30.32-bit lower quality157 GB75–133
MiniMax M3428B · 23B active · Intelligence 29.22-bit lower quality134 GB75–133
Nemotron 3 Ultra 550B A55B550B · 55B active · Intelligence 22.92-bit lower quality172 GB31–56
Qwen3.5 397B A17B397B · 17B active · Intelligence 18.43-bit lower quality174 GB73–130
Llama 4 Maverick 17B-128E Instruct400B · 17B active · Intelligence 9.33-bit lower quality175 GB73–130

Runs, but slowly 2

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 GB4–7
Devstral 2 (123B)123B8-bit131 GB4–7

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 GB42–75

Too big 9

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

Other Macs

Every Mac with 256 GB, from any chip.

The M5 Ultra with more or less memory:

Neighbouring chips:

Head to head:

Against PCs and graphics cards:

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

Questions

What is the best LLM for an M5 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 110–196 tokens a second. The biggest that runs well is MiMo V2.6 Flash.
Can an M5 Ultra with 256 GB run Llama 3.3 70B?
Yes: 4-bit, at about 14–24 tokens a second.
How fast is gpt-oss-20b on an M5 Ultra?
About 284–505 tokens a second (native 4.25-bit), from the chip’s 1,200 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 Ultra with 256 GB run? Models and speed