Apple M3 Max (40-core GPU) × Llama 3.2 1B Instruct
Llama 3.2 1B Instruct on Apple M3 Max (40-core GPU)
153best gen tok/s
3,403best prompt tok/s
4,352max context tested
7results
Measured · all numbers below come from the linked sources
llama.cpp
| Quant | Gen tok/s | Prompt tok/s | Source | Context tested | TTFT | Build |
|---|---|---|---|---|---|---|
| Q4_K - Medium | 153 | 3,348 | LocalScore ↗ | 4,352 | 357 ms | llamafile 0.9.2 |
| Q4_K - Medium | 150 | 3,395 | LocalScore ↗ | 512+128 | 348 ms | llamafile 0.9.2 |
| Q4_K - Medium | 148 | 3,403 | LocalScore ↗ | 512+128 | 350 ms | llamafile 0.9.2 |
| Q4_K - Medium | 148 | 3,209 | LocalScore ↗ | 4,352 | 363 ms | llamafile 0.9.2 |
| Q4_K - Medium | 146 | 3,292 | LocalScore ↗ | 512+128 | 358 ms | llamafile 0.9.2 |
| Q4_K - Medium | 109 | 2,585 | LocalScore ↗ | 512+128 | 508 ms | llamafile 0.9.2 |
| Q4_K - Medium | 105 | 2,458 | LocalScore ↗ | 512+128 | 533 ms | llamafile 0.9.2 |
Ollama
No measured Ollama results yet. Submit yours →
MLX
No measured MLX results yet. Submit yours →
"Context tested" is the largest prompt+generation length actually benchmarked (tokens), not a memory-fit estimate.
What else can Apple M3 Max (40-core GPU) run? →
Includes results from LocalScore, a Mozilla Builders project by cjpais, mirrored in the cn0303/local-llm-speed-benchmarks dataset. LocalScore runs use llamafile, which is built on llama.cpp.