Skip to content

What can the M3 Ultra 256 GB run?

What fits on the M3 Ultra 256 GB
Text models: 80Visual models: 2Strongest first
Memory
256 GB unified
Bandwidth
819 GB/s
Maker
Apple

Open the calculator on this machineApple M3 Ultra in our inventory

Text models that run on it

Text models that run on the M3 Ultra 256 GB, best first
ModelBest fitContextSpeedStrongest at
DeepSeek V4.1 Flash1.6-bit1M82 tok/sChat, Coding, Agents, Docs
GLM-5.31.6-bit384K38 tok/sChat, Coding, Agents, Docs
GLM-5.3 Flash3-bit1M43 tok/sChat, Coding, Agents, Docs
Qwen3.8 Flash-Next8-bit136K53 tok/sChat, Coding, Agents
Muse Spark Open 816B A42B1.6-bit352K40 tok/sChat, Docs
MiMo V2.6 FlashFP4 + FP8928K37 tok/sCoding, Agents
MiniMax M32-bit256K41 tok/sDocs
DeepSeek V4 FlashFP4 + FP81M24 tok/sDocs
Hy34-bit48K28 tok/sChat, Coding, Agents, Docs
Nemotron 3 Ultra 550B A55B1.6-bit256K27 tok/sChat, Coding, Docs
Qwen3.5 397B A17B2-bit256K48 tok/sAgents, Docs
MiniMax M2.76-bit28K42 tok/sCoding, Agents
DeepSeek V3.21.6-bit160K44 tok/sChat, Coding, Agents, Docs
Mistral Medium 3.5 128BFP8160K3.9 tok/sChat, Coding, Agents, Docs
Qwen3.5 122B A10B8-bit256K37 tok/sChat, Coding, Agents, Docs
Nemotron 3 Super 120B A12B8-bit256K27 tok/sChat, Coding, Agents, Docs
Nemotron 3.5 Bolt 74B A9B8-bit256K33 tok/sChat, Coding, Agents, Docs
Mistral Small 4 119BFP8256K57 tok/sChat, Docs
gpt-oss 120BMXFP4128K84 tok/sChat, Coding
Qwen3 Coder Next 80B8-bit256K78 tok/sCoding, Agents
Llama 3.3 70B8-bit128K6.9 tok/sChat, Coding, Agents, Docs
Inkling1.6-bit4K40 tok/sText and images
DeepSeek R11.6-bit160K42 tok/sNot rated
DeepSeek V31.6-bit160K42 tok/sNot rated
GLM 51.6-bit198K37 tok/sNot rated
GLM 4.5 Air8-bit128K28 tok/sNot rated
Inkling Small5-bit4K47 tok/sText and images
MiniMax M2.56-bit28K38 tok/sNot rated
GLM 5.11.6-bit198K37 tok/sNot rated
Laguna M.16-bit32K22 tok/sNot rated
Laguna S 2.18-bit1M42 tok/sNot rated
Qwen3 VL 235B A22B Instruct5-bit136K23 tok/sText and images
GLM 5.21.6-bit384K37 tok/sNot rated
Qwen3 235B A22B5-bit40K23 tok/sNot rated
Ornith 1.5 122B8-bit256K31 tok/sNot rated
Ornith 1.5 397B2-bit256K39 tok/sNot rated
Ornith 1.0 397B3-bit24K40 tok/sNot rated
Step 3.5 Flash6-bit72K48 tok/sNot rated
DeepSeek V3 03241.6-bit160K42 tok/sNot rated
Qwen2.5 72B Instruct8-bit32K6.7 tok/sNot rated
Qwen3 Next 80B A3B Instruct8-bit256K78 tok/sNot rated
DeepSeek R1 05281.6-bit160K42 tok/sNot rated
GLM 4.53-bit48K22 tok/sNot rated
Llama 4 Scout 17B 16E Instruct8-bit384K23 tok/sText and images
MiniMax M26-bit28K38 tok/sNot rated
MiMo V2.54-bit256K38 tok/sNot rated
Qwen2.5 VL 72B Instruct8-bit125K6.6 tok/sText and images
Solar Open 100B8-bit128K32 tok/sNot rated
GLM 5.3 Flash BF163-bit1M43 tok/sText and images
Qwen3 VL 235B A22B Thinking5-bit136K23 tok/sText and images
Qwen3 235B A22B Instruct 25075-bit136K23 tok/sNot rated
NVIDIA Nemotron 3.5 Super VL 120B A12B BF168-bit1M27 tok/sText and images
Hy3 Preview4-bit40K29 tok/sNot rated
Qwen3 Coder 480B A35B Instruct2-bit56K30 tok/sCoding
GLM 4.73-bit48K22 tok/sNot rated
Kimi Linear 48B A3B Instruct8-bit4K102 tok/sNot rated
Qwen3 Next 80B A3B Thinking8-bit256K78 tok/sNot rated
GLM 5.3 BF161.6-bit384K37 tok/sNot rated
GLM 4.63-bit52K23 tok/sNot rated
DeepSeek R1 Distill Llama 70B8-bit128K6.9 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 80 are shown.

Image and video models

Image and video models that run on the M3 Ultra 256 GB
ModelBest fitPer outputStrongest at
MiniMax H316-bit35 minVideo
FLUX.2 [dev]16-bit9.3 minImages

Only just too big

These need up to a quarter more memory than the M3 Ultra 256 GB has. Spilling into system RAM or a lower precision can get them running, slowly.

Questions

What is the best LLM for the M3 Ultra 256 GB?

Of the open models we track, the strongest that run well on the M3 Ultra 256 GB are DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3 Flash. 80 text models fit it in all.

Calculated from published specs, not benchmarked. All machines