Apple M4 Max (40-core GPU) × Llama 3.2 1B Instruct
Llama 3.2 1B Instruct on Apple M4 Max (40-core GPU)
191best gen tok/s
4,072best prompt tok/s
4,352max context tested
26results
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 | 191 | 3,879 | LocalScore ↗ | 512+128 | 301 ms | llamafile 0.9.2 |
| Q4_K - Medium | 190 | 3,888 | LocalScore ↗ | 512+128 | 304 ms | llamafile 0.9.2 |
| Q4_K - Medium | 190 | 3,952 | LocalScore ↗ | 512+128 | 297 ms | llamafile 0.9.2 |
| Q4_K - Medium | 186 | 3,699 | LocalScore ↗ | 512+128 | 311 ms | llamafile 0.9.2 |
| Q4_K - Medium | 185 | 3,883 | LocalScore ↗ | 512+128 | 305 ms | llamafile 0.9.2 |
| Q4_K - Medium | 184 | 3,893 | LocalScore ↗ | 512+128 | 301 ms | llamafile 0.9.1 |
| Q4_K - Medium | 184 | 3,776 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 183 | 3,878 | LocalScore ↗ | 512+128 | 301 ms | llamafile 0.9.2 |
| Q4_K - Medium | 182 | 3,746 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 181 | 4,072 | LocalScore ↗ | 512+128 | 295 ms | llamafile 0.9.2 |
| Q4_K - Medium | 181 | 3,877 | LocalScore ↗ | 512+128 | 303 ms | llamafile 0.9.2 |
| Q4_K - Medium | 181 | 3,925 | LocalScore ↗ | 512+128 | 302 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,769 | LocalScore ↗ | 512+128 | 309 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,827 | LocalScore ↗ | 512+128 | 307 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,751 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,918 | LocalScore ↗ | 512+128 | 298 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,858 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 180 | 3,608 | LocalScore ↗ | 512+128 | 314 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,814 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,654 | LocalScore ↗ | 4,352 | 318 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,799 | LocalScore ↗ | 512+128 | 305 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,731 | LocalScore ↗ | 512+128 | 309 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,756 | LocalScore ↗ | 512+128 | 308 ms | llamafile 0.9.2 |
| Q4_K - Medium | 179 | 3,896 | LocalScore ↗ | 512+128 | 302 ms | llamafile 0.9.2 |
| Q4_K - Medium | 178 | 3,658 | LocalScore ↗ | 512+128 | 312 ms | llamafile 0.9.2 |
| Q4_K - Medium | 177 | 3,800 | LocalScore ↗ | 512+128 | 307 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 M4 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.