Apple M2 Pro (19-core GPU)

12measured results

104best 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	104	1,508	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	100	1,473	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	99.6	1,485	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	99.5	1,458	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	99.3	1,495	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	98.0	1,477	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	Q4_0	38.9	341	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Meta Llama 3.1 8B Instruct	Q4_K - Medium	26.3	262	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Meta Llama 3.1 8B Instruct	Q4_K - Medium	26.3	262	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	Q8_0	23.0	345	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Qwen2.5 14B Instruct	Q4_K - Medium	14.1	137	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	F16	13.1	384	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	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.