Below you will find a very handy chart to compare TPS (token per second) in among of different Thunderbolt Mac minis. The TPS column uses published Llama 2 7B, Q4_0, GPU-accelerated token-generation benchmarks—the tg128 test, not prompt-processing throughput.
Mac mini | CPU | GPU | Unified memory | Memory | TPS for |
|---|---|---|---|---|---|
M4 Pro — 2024 | 12 | 16 | 24 / 48 / 64GB | 273 GB/s | 49.6 TPS |
M4 Pro — 2024 | 14 | 20 | 24 / 48 / 64GB | 273 GB/s | 50.7 TPS |
M5 Pro — 2026 | 15 | 16 | 24 / 48 / 64GB | 307 GB/s | 63.9 TPS |
M5 Pro — 2026 | 18 | 20 | 24 / 48 / 64GB | 307 GB/s | 66.3 TPS |