Apple M3 Max (40-core GPU) × Qwen2.5 14B Instruct
Qwen2.5 14B Instruct on Apple M3 Max (40-core GPU)
25.5best gen tok/s
303best prompt tok/s
4,352max context tested
6results
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 | 25.5 | 303 | LocalScore ↗ | 512+128 | 4,113 ms | llamafile 0.9.2 |
| Q4_K - Medium | 25.5 | 302 | LocalScore ↗ | 512+128 | 4,029 ms | llamafile 0.9.2 |
| Q4_K - Medium | 24.0 | 296 | LocalScore ↗ | 4,352 | 4,077 ms | llamafile 0.9.2 |
| Q4_K - Medium | 23.6 | 287 | LocalScore ↗ | 4,352 | 4,269 ms | llamafile 0.9.2 |
| Q4_K - Medium | 13.9 | 202 | LocalScore ↗ | 512+128 | 6,648 ms | llamafile 0.9.2 |
| Q4_K - Medium | 13.7 | 200 | LocalScore ↗ | 512+128 | 6,666 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.