How we got this number
Each generated token reads the whole model from memory once: 55 GB. The Mac Studio M5 Ultra (96GB) moves 1228.8 GB/s, so the ceiling is 22 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 3.2 Vision 90B
Other models on the Mac Studio M5 Ultra (96GB)
| Nemotron 3.5 lightning 30B (A3B) | Fast | 120+ tok/s |
| Qwen 3 30B (A3B) | Fast | 120+ tok/s |
| Gemma 4 26B (A4B) | Fast | 120+ tok/s |
| Qwen3.5 35B (A3B) | Fast | 120+ tok/s |
| gpt-oss 20B | Fast | 120+ tok/s |
| gpt-oss 120B | Fast | 120+ tok/s |