How we got this number
Each generated token reads the whole model from memory once: 9.1 GB. The Mac mini M6 (16GB) moves 170 GB/s, so the ceiling is 19 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 Phi 4 14B
| Mac Studio M5 Ultra (96GB) | 108 tok/s | Check price → |
| Mac Studio M5 Ultra (256GB) | 108 tok/s | Check price → |
| Mac Studio M5 Max (40-core GPU, 64GB) | 54 tok/s | Check price → |
| Mac mini M5 Pro (24GB) | 27 tok/s | Check price → |
Other models on the Mac mini M6 (16GB)
| lfm 2.5 8B (A1B) | Fast | 111 tok/s |
| Qwen2.5 3B | Fast | 72 tok/s |
| Qwen2.5 Coder 3B | Fast | 72 tok/s |
| Qwen3.5 2B | Fast | 72 tok/s |
| Qwen 3-VL 2B | Fast | 72 tok/s |
| Llama 3.2 3B | Fast | 68 tok/s |