How we got this number
Each generated token reads the whole model from memory once: 21 GB. The Mac mini M4 (16GB) moves 120 GB/s, so the ceiling is 5.7 tokens/sec. Real machines reach a fraction of that ceiling; we use 80%, calibrated against a measured Mac run. Send us your real numbers and we'll replace the estimate.
Faster machines for Qwen2.5-VL 32B
| Mac Studio M5 Ultra (96GB) | 47 tok/s | Check price → |
| Mac Studio M5 Ultra (256GB) | 47 tok/s | Check price → |
| Mac Studio M5 Max (40-core GPU, 64GB) | 23 tok/s | Check price → |
| Mac mini M5 Pro (64GB) | 12 tok/s | Check price → |
Other models on the Mac mini M4 (16GB)
| lfm 2.5 8B (A1B) | Fast | 78 tok/s |
| Mistral 7B | Fast | 22 tok/s |
| Qwen2 7B | Fast | 22 tok/s |
| starCoder 2 7B | Fast | 22 tok/s |
| OLMo 2 7B | Fast | 21 tok/s |
| falcon 3 7B | Fast | 21 tok/s |