Written by Jakub Rusinowski · Last updated April 28, 2025
Yes
Yes — Qwen 3 235B-A22B (MoE) at Q5_K_M needs about 168.9 GB of the 192 GB usable on Mac Studio (M3 Ultra, 256 GB) (~23.1 GB spare), at ~25.3 tok/s (estimated), with room for about 65,536 tokens of context.
Confidence: high · Recommended quantization: Q5_K_M · Estimated speed: ~25.3 tok/s
| Usable memory for models | 192 GB |
| Memory bandwidth | 819 GB/s |
| Form factor | Desktop |
| Operating system | macOS |
| Memory upgradeable | No — soldered |
| Quant | Memory needed | Fits 192 GB? | Max context | Est. speed | Download |
|---|---|---|---|---|---|
| F16 | 472.4 GB | ✗ No | — | — | 470 GB |
| Q8_0 | 252.1 GB | ✗ No | — | — | 249.7 GB |
| Q6_K | 195.1 GB | ✗ No | — | — | 192.7 GB |
| Q5_K_M | 168.9 GB | ✓ Yes | 64K | ~25.3 tok/s | 166.6 GB |
| Q4_K_M | 144.3 GB | ✓ Yes | 64K | ~29.1 tok/s | 141.9 GB |
| Q3_K_M | 102.5 GB | ✓ Yes | 64K | ~39 tok/s | 100.2 GB |
| Q2_K | 79.6 GB | ✓ Yes | 64K | ~47.9 tok/s | 77.3 GB |
| Variant | Needs at Q4_K_M | Fits? | Est. speed |
|---|---|---|---|
| Qwen 3 235B-A22B (MoE) | 144.3 GB | ✓ Fits | ~29.1 tok/s |
| Qwen 3 32B | 22.8 GB | ✓ Fits | ~20.2 tok/s |
| Qwen 3 30B-A3B (MoE) | 20 GB | ✓ Fits | ~120 tok/s |
| Qwen 3 14B | 11.1 GB | ✓ Fits | ~41 tok/s |
| Qwen 3 8B | 7 GB | ✓ Fits | ~65 tok/s |
Ollama or LM Studio (Metal) — MLX for the fastest Apple-native throughput
Yes — Qwen 3 235B-A22B (MoE) at Q5_K_M needs about 168.9 GB of the 192 GB usable on Mac Studio (M3 Ultra, 256 GB) (~23.1 GB spare), at ~25.3 tok/s (estimated), with room for about 65,536 tokens of context.
Q5_K_M — it needs about 168.9 GB of the 192 GB available, downloads as roughly 166.6 GB, and runs at an estimated 25.3 tokens/sec with up to 64K of context.
Nothing binding — the model fits with headroom and generates at a usable speed on this hardware.
Ollama or LM Studio (Metal) — MLX for the fastest Apple-native throughput