Skip to content

What can the M3 Ultra 96 GB run?

What fits on the M3 Ultra 96 GB
Text models: 112Visual models: 14Strongest first
Memory
96 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 96 GB, best first
ModelBest fitContextSpeedStrongest at
Qwen3.8 Flash-Next2-bit192K90 tok/sChat, Coding, Agents
Qwen3.8 27B8-bit256K17 tok/sChat, Coding, Agents, Docs
Muse Glimmer 30B8-bit128K16 tok/sDocs
Mistral Medium 3.5 128B3-bit24K8.1 tok/sChat, Coding, Agents, Docs
Qwen3.5 122B A10B3-bit256K62 tok/sChat, Coding, Agents, Docs
Nemotron 3 Super 120B A12B3-bit256K42 tok/sChat, Coding, Agents, Docs
Nemotron 3.5 Bolt 74B A9B6-bit256K38 tok/sChat, Coding, Agents, Docs
Qwen3.6 35B A3B8-bit256K85 tok/sChat, Coding, Agents, Docs
Gemma 4 31B8-bit256K15 tok/sChat, Coding, Agents, Docs
Gemma 4 26B A4B8-bit256K85 tok/sChat, Coding, Agents, Docs
GLM-4.7 Flash8-bit198K80 tok/sChat, Coding, Agents
Mistral Small 4 119B4-bit16K80 tok/sChat, Docs
gpt-oss 120BMXFP4128K84 tok/sChat, Coding
Qwen3 Coder Next 80B6-bit256K87 tok/sCoding, Agents
Nemotron 3.5 Lightning 30B A3B8-bit256K75 tok/sAgents, Docs
Devstral Small 2 24BFP8256K20 tok/sCoding
LFM2.5 24B A2B8-bit128K104 tok/sAgents
Mistral Small 3.2 24B8-bit128K20 tok/sChat, Coding, Agents
Ministral 3 14B8-bit256K34 tok/sCoding, Agents, Docs
Llama 3.3 70B6-bit44K8.8 tok/sChat, Coding, Agents, Docs
Olmo 3.1 32B8-bit64K15 tok/sChat, Coding, Agents, Docs
Qwen3 14B8-bit40K32 tok/sNot rated
Qwen3 32B8-bit40K15 tok/sNot rated
Qwen3.6 27B8-bit256K17 tok/sText and images
Qwen2.5 Coder 14B Instruct8-bit32K31 tok/sCoding
Qwen3 Coder 30B A3B Instruct8-bit256K79 tok/sCoding
Qwen3.5 27B8-bit256K17 tok/sText and images
Qwen3.5 35B A3B8-bit256K85 tok/sText and images
NVIDIA Nemotron 3 Nano 30B A3B BF168-bit256K75 tok/sNot rated
GLM 4.5 Air4-bit28K43 tok/sNot rated
Qwen3 30B A3B8-bit40K79 tok/sNot rated
Qwen3 30B A3B Instruct 25078-bit256K79 tok/sNot rated
Laguna XS.28-bit256K86 tok/sNot rated
OTel 2.0 LLM 31B IT8-bit256K15 tok/sNot rated
Qwen2.5 32B Instruct8-bit32K15 tok/sNot rated
MiniCPM V 4.7 35B A3B8-bit256K85 tok/sText and images
Qwen2.5 14B Instruct8-bit32K31 tok/sNot rated
Laguna XS 2.18-bit256K86 tok/sNot rated
Diffusiongemma 26B A4B IT8-bit256K82 tok/sText and images
Qwen2.5 VL 32B Instruct8-bit125K14 tok/sText and images
Qwen2.5 Coder 32B Instruct8-bit32K15 tok/sCoding
Laguna S 2.14-bit28K61 tok/sNot rated
Qwen3 VL 30B A3B Instruct8-bit256K79 tok/sText and images
DeepSeek R1 Distill Qwen 32B8-bit128K15 tok/sNot rated
Gemma 3 27B IT8-bit128K18 tok/sText and images
Ornith 1.5 35B A3B8-bit256K85 tok/sNot rated
Ornith 1.5 122B3-bit256K53 tok/sNot rated
GPT Neox 20B8-bit2K23 tok/sNot rated
Phi 48-bit16K32 tok/sNot rated
Qwen3 VL 32B Instruct8-bit136K14 tok/sText and images
Pixtral 12B8-bit352K37 tok/sText and images
DeepSeek R1 Distill Qwen 14B8-bit128K31 tok/sNot rated
Qwen2.5 14B Instruct 1M8-bit256K31 tok/sNot rated
Qwen2.5 72B Instruct6-bit32K8.6 tok/sNot rated
Qwen3 Next 80B A3B Instruct6-bit232K87 tok/sNot rated
Llama 4 Scout 17B 16E Instruct4-bit32K37 tok/sText and images
Llama Guard 4 12B8-bit288K39 tok/sText and images
Qwen2.5 VL 72B Instruct6-bit36K8.5 tok/sText and images
Qwen1.5 7B8-bit32K60 tok/sNot rated
Solar Open 100B4-bit48K49 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 112 are shown.

Image and video models

Image and video models that run on the M3 Ultra 96 GB
ModelBest fitPer outputStrongest at
MiniMax H38-bit35 minVideo
LTX-2.516-bit3.8 minVideo
FLUX.1 [dev]16-bit66 sImages
Qwen-Image16-bit21 minImages
FLUX.2 [dev]8-bit9.3 minImages
Qwen-Image 2.116-bit9.1 minImages
Wan2.2 I2V A14B16-bit3.1 hVideo
FLUX.2 [klein] 9B16-bit16 sImages
Wan2.1 I2V 14B16-bit3.1 hVideo
Wan 2.1 T2V-14B16-bit3.9 hVideo
Krea 2 Turbo16-bit66 sImages
Qwen Image 251216-bit21 minImages
Wan 2.2 A14B16-bit3.1 hVideo
HunyuanVideo 1.516-bit3.8 hVideo

Only just too big

Questions

What is the best LLM for the M3 Ultra 96 GB?

Of the open models we track, the strongest that run well on the M3 Ultra 96 GB are Qwen3.8 Flash-Next, Qwen3.8 27B, Muse Glimmer 30B. 112 text models fit it in all.

Calculated from published specs, not benchmarked. All machines