Written by Jakub Rusinowski · Last updated July 12, 2026
Best all-round pick: Qwen 3 235B-A22B (MoE)
The Apple M2 Ultra has 192 GB of unified memory, of which about 144 GB is available to a model. The largest model it can hold is Qwen 3 235B-A22B (MoE) (235B, 142.7 GB at Q4_K_M).
| Model | Score | Memory at Q4_K_M | Context | Licence |
|---|---|---|---|---|
| Qwen 3 235B-A22B (MoE) | 97 | 142.7 GB | 125K | Apache 2.0 |
| GPT-oss 120B | 95.1 | 73.3 GB | 125K | Apache-2.0 |
| MiniMax M3 230B-A10B | 94.4 | 139.7 GB | 977K | Modified MIT (attribution required) |
| MiniMax M2.5 230B | 94.2 | 139.7 GB | 977K | Modified MIT (attribution required) |
| Nemotron 70B Instruct | 94.1 | 43.4 GB | 125K | Llama Community |
| Qwen 3.5 122B-A10B (MoE) | 93.3 | 74.5 GB | 125K | Apache 2.0 |
| Workload | Recommended on this GPU |
|---|---|
| Coding | Devstral-2 123B (97.8) MiniMax M3 230B-A10B (97.7) MiniMax M2.5 230B (97.4) |
| General assistant | Qwen 3 235B-A22B (MoE) (97) GPT-oss 120B (95.1) MiniMax M3 230B-A10B (94.4) |
| Reasoning | Qwen 3 235B-A22B (MoE) (100) GLM-5.1 72B (98) Qwen 3.5 122B-A10B (MoE) (96.4) |
| RAG | MiniMax M3 230B-A10B (96.3) MiniMax M2.5 230B (96.2) Llama 4 Scout 17B (93.1) |
| Agents | MiniMax M2.7 230B-A10B (89.1) Ternary Bonsai 27B (88.8) Qwen 3.5 122B-A10B (MoE) (86.9) |
| Vision | MiniMax M3-VL (99.4) Qwen 3.5 72B (96.4) Llama 3.2 90B Vision Instruct (94.7) |
Qwen 3 235B-A22B (MoE) is the strongest all-round pick that fits its 144 GB.
Qwen 3 235B-A22B (MoE) — 235B parameters, needing 142.7 GB at Q4_K_M.
About 144 GB of its 192 GB, because macOS reserves a share of unified memory for the system.