How we got this number
Each generated token reads the whole model from memory once: 39 GB. The Mac mini M6 (32GB) moves 170 GB/s, so the ceiling is 4.4 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 Llama 2 70B
| Mac Studio M5 Ultra (96GB) | 25 tok/s | Check price → |
| Mac Studio M5 Ultra (256GB) | 25 tok/s | Check price → |
| Mac Studio M5 Max (40-core GPU, 64GB) | 13 tok/s | Check price → |
| Mac mini M5 Pro (64GB) | 6.3 tok/s | Check price → |
Other models on the Mac mini M6 (32GB)
| Qwen 3 30B (A3B) | Fast | 40 tok/s |
| Qwen3 Coder 30B (A3B) | Fast | 40 tok/s |
| gpt-oss 20B | Fast | 33 tok/s |
| Gemma 4 26B (A4B) | Fast | 27 tok/s |
| Qwen 3-VL 30B (A3B) | Fast | 23 tok/s |
| Qwen3.5 35B (A3B) | Fast | 21 tok/s |