All devices
64 / 128 / 192 GB unified · 800 GB/s · Metal
64 / 128 / 192 GB unified · 800 GB/s · MetalAI models for M2 Ultra
Here are the best open models for each job on the M2 Ultra, and every model's grade below.
Your hardware
…
- Memory
- GB
- Bandwidth
- GB/s
- Context
Most intelligent on your …
Intelligence (ECI) of the most intelligent models that load · colour shows how well they run
Expected speed
Tokens per second on your … for the most intelligent models that run well
Models available with more memory
Open models that run well (grade B or better) at each memory size
Scores from Oct 8, 2026 · 0 new models in the last two months
Every model on the M2 Ultra
Q4_K_M, 8K context, 64 GB of memory. Change your setup above for exact numbers.
| Grade | Model | Memory | Speed |
|---|---|---|---|
| SExcellent |  Kimi Linear 48B A3B 49.1B | 28.4 GB | ~127 tok/s |
| SExcellent |  Qwen3.6 35B A3B 36B | 21.2 GB | ~126 tok/s |
| SExcellent | Qwen3.5 35B A3B 36B | 21 GB | ~126 tok/s |
| SExcellent |  Nemotron 3.5 Lightning 30B A3B 31.6B | 24.1 GB | ~116 tok/s |
| SExcellent |  GLM 4.7 Flash 31.2B | 17.8 GB | ~125 tok/s |
| SExcellent | Qwen3 30B A3B Thinking 2507 30.5B | 18.3 GB | ~116 tok/s |
| SExcellent | Qwen3 30B A3B Instruct 2507 30.5B | 18.3 GB | ~116 tok/s |
| SExcellent |  North Mini Code 1.0 30.5B | 18.2 GB | ~119 tok/s |
| SExcellent |  Gemma 4 26B A4B 25.8B | 16.5 GB | ~110 tok/s |
| SExcellent |  gpt-oss 20b 20.9B | 11.3 GB | ~122 tok/s |
| SExcellent |  Phi-4 14B 14.7B | 10.1 GB | ~54 tok/s |
| SExcellent |  Ministral 3 14B 14B | 9.2 GB | ~58 tok/s |
| SExcellent | Gemma 4 12B 12B | 7.4 GB | ~65 tok/s |
| SExcellent | Qwen3.5 9B 9.7B | 5.8 GB | ~76 tok/s |
| SExcellent | Ministral 3 8B 8.9B | 6.2 GB | ~76 tok/s |
| SExcellent |  Granite 4.2 8B 8.8B | 6.7 GB | ~73 tok/s |
| SExcellent |  Llama 3.1 8B Instruct 8B | 5.9 GB | ~79 tok/s |
| SExcellent | Gemma 4 E4B 8B | 5.1 GB | ~82 tok/s |
| SExcellent |  Olmo 3 7B Instruct 7.3B | 7 GB | ~70 tok/s |
| SExcellent | Gemma 4 E2B 5.1B | 3.3 GB | ~104 tok/s |
| SExcellent | Qwen3.5 4B 4.7B | 3.1 GB | ~109 tok/s |
| SExcellent | Ministral 3 3B 3.9B | 3.1 GB | ~113 tok/s |
| SExcellent | Granite 4.2 3B 3.7B | 3.1 GB | ~112 tok/s |
| SExcellent | Llama 3.2 3B Instruct 3.2B | 3.1 GB | ~115 tok/s |
| SExcellent |  SmolLM3 3B 3.1B | 2.6 GB | ~120 tok/s |
| SExcellent |  LFM2.5 2.6B 2.7B | 2 GB | ~130 tok/s |
| SExcellent | Qwen3.5 2B 2.3B | 1.6 GB | ~140 tok/s |
| SExcellent | Qwen3.5 0.8B 0.9B | 0.9 GB | ~163 tok/s |
| AGood | Olmo 3.1 32B Think 32.2B | 19.7 GB | ~30 tok/s |
| AGood | Gemma 4 31B 31.3B | 18.8 GB | ~32 tok/s |
| AGood | Granite 4.2 30B 29.3B | 19.1 GB | ~32 tok/s |
| AGood | Qwen3.8 27B 27.8B | 17 GB | ~34 tok/s |
| AGood | Qwen3.6 27B 27.8B | 16.5 GB | ~35 tok/s |
| AGood | Devstral Small 2 24B 24B | 14.9 GB | ~39 tok/s |
| AGood | Magistral Small 2509 24B | 14.9 GB | ~39 tok/s |
| AGood | Mistral Small 3.2 24B 24B | 14.9 GB | ~39 tok/s |
| AGood | Magistral Small 2506 23.6B | 14.9 GB | ~39 tok/s |
| BUsable | Qwen3 Coder Next 79.7B | 45.7 GB | ~126 tok/s |
| BUsable | Llama 3.3 70B Instruct 70.6B | 42.4 GB | ~16 tok/s |
35 models do not fit on the M2 Ultra
| Grade | Model | Memory | Speed |
|---|---|---|---|
| FDoesn't fit | Kimi K3 2.8T | 1,584.5 GB | — |
| FDoesn't fit |  DeepSeek V4 Pro 0813 1.7T | 886.8 GB | — |
| FDoesn't fit | DeepSeek V4 Pro 1.6T | 886.8 GB | — |
| FDoesn't fit | Kimi K2.6 1T | 578.8 GB | — |
| FDoesn't fit | Kimi K2.7 Code 1T | 578.8 GB | — |
| FDoesn't fit | Kimi K2.5 1T | 579.4 GB | — |
| FDoesn't fit | Kimi K2 Thinking 1T | 579.4 GB | — |
| FDoesn't fit | Kimi K2 Instruct 1T | 579 GB | — |
| FDoesn't fit | tInkling 952.4B | 534 GB | — |
| FDoesn't fit | DeepSeek V4.1 Flash 763.2B | 435.8 GB | — |
| FDoesn't fit | GLM 5.1 753.9B | 429.2 GB | — |
| FDoesn't fit | GLM 5 753.9B | 425.7 GB | — |
| FDoesn't fit | GLM 5.2 753.3B | 424.5 GB | — |
| FDoesn't fit | DeepSeek V3.2 Exp 685.4B | 391.5 GB | — |
| FDoesn't fit | DeepSeek V3.2 685.4B | 378.5 GB | — |
| FDoesn't fit | DeepSeek V3.1 684.5B | 378.4 GB | — |
| FDoesn't fit | Nemotron 3 Ultra 550B A55B 560.5B | 321.8 GB | — |
| FDoesn't fit |  MiniMax M3 427B | 244.6 GB | — |
| FDoesn't fit | Qwen3.5 397B A17B 403.4B | 227.9 GB | — |
| FDoesn't fit | GLM 4.7 358.3B | 204.8 GB | — |
| FDoesn't fit | GLM 5.3 Flash 321.3B | 176.1 GB | — |
| FDoesn't fit | DeepSeek V4 Flash 0731 304.2B | 161 GB | — |
| FDoesn't fit | DeepSeek V4 Flash 290.9B | 166.7 GB | — |
| FDoesn't fit | tInkling Small 266B | 149 GB | — |
| FDoesn't fit | Qwen3 235B A22B Thinking 2507 235.1B | 134.2 GB | — |
| FDoesn't fit | Qwen3 235B A22B Instruct 2507 235.1B | 134.2 GB | — |
| FDoesn't fit | MiniMax M2.5 228.7B | 131.1 GB | — |
| FDoesn't fit | MiniMax M2.7 228.7B | 131.5 GB | — |
| FDoesn't fit | Command A+ 218.8B | 126.7 GB | — |
| FDoesn't fit | Qwen3.8 Flash Next 180B | 111.9 GB | — |
| FDoesn't fit | Mistral Medium 3.5 128B 127.7B | 72.8 GB | — |
| FDoesn't fit | Qwen3.5 122B A10B 125.1B | 71.8 GB | — |
| FDoesn't fit | Mistral Small 4 119B 119.4B | 68.1 GB | — |
| FDoesn't fit | gpt-oss 120b 116.8B | 59 GB | — |
| FDoesn't fit | Llama 4 Scout 17B 16E 108.6B | 62.7 GB | — |
Next steps
Best models that run on M2 UltraThe intelligence ranking, filtered to what this device runs.Tier list for M2 UltraEvery model grouped by grade, ready to share as an image.M2 Ultra vs M5 UltraWhat the next step up would change, model by model.