Apple M1 Ultra (64-core GPU)

9measured results

154best gen tok/s*

128 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	154	2,998	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 3.2 1B Instruct	Q4_K - Medium	148	2,956	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	Q4_0	83.7	1,030	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Llama 2 7B	Q8_0	59.9	1,043	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Meta Llama 3.1 8B Instruct	Q4_K - Medium	54.4	675	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Meta Llama 3.1 8B Instruct	Q4_K - Medium	54.2	660	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Llama 2 7B	F16	37.0	1,169	llama.cpp (Metal) ↗	512+128	llama.cpp Metal	Details →
Qwen2.5 14B Instruct	Q4_K - Medium	33.3	377	LocalScore ↗	512+128	llamafile 0.9.2	Details →
Qwen2.5 14B Instruct	Q4_K - Medium	31.5	366	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.