Apple M2 Max (30-core GPU)

9measured results

135best gen tok/s*

32 GBreported memory

*Headline excludes tiny/test models (under 0.6B params or TinyStories-style). Open in rig checker →

llama.cpp

Model	Quant	Gen tok/s	Prompt tok/s	Source	Context tested	Build
Llama 3.2 1B Instruct	Q4_K - Medium	135	2,105	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	129	2,038	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	123	2,049	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	123	1,935	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	Q4_0	61.0	538	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Llama 2 7B	Q8_0	40.0	540	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Meta Llama 3.1 8B Instruct	Q4_K - Medium	31.2	325	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	F16	24.2	600	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Qwen2.5 14B Instruct	Q4_K - Medium	14.5	149	LocalScore ↗	512+128	llamafile 0.9.2	Details →

Ollama

No measured Ollama results yet. Submit yours →

MLX

No measured MLX results yet. Submit yours →

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.