How we got this number
Each generated token reads the whole model from memory once: 40 GB. The Mac mini M2 (16GB) moves 100 GB/s, so the ceiling is 2.5 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 3 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) | 12 tok/s | Check price → |
| Mac mini M5 Pro (64GB) | 6.1 tok/s | Check price → |
Other models on the Mac mini M2 (16GB)
| Gemma 3 12B | Usable, but slow | 9.9 tok/s |
| translateGemma 12B | Usable, but slow | 9.9 tok/s |
| Qwen 14B | Usable, but slow | 9.8 tok/s |
| OLMo 2 13B | Usable, but slow | 9.5 tok/s |
| DeepSeek Coder v2 16B | Usable, but slow | 9.0 tok/s |
| DeepSeek-R1 14B | Usable, but slow | 8.9 tok/s |