Hardware /

Apple M4 Max (40-core GPU)

70measured results
191best gen tok/s*
64 GBreported memory

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

llama.cpp

ModelQuantGen tok/sPrompt tok/sSourceContext testedBuild
Llama 3.2 1B InstructQ4_K - Medium1913,879LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1903,888LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1903,952LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1863,699LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1853,883LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1843,893LocalScore ↗512+128llamafile 0.9.1Details →
Llama 3.2 1B InstructQ4_K - Medium1843,776LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1833,878LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1823,746LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1814,072LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1813,877LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1813,925LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,769LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,827LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,751LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,918LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,858LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1803,608LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,814LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,654LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,799LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,731LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,756LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1793,896LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1783,658LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium1773,800LocalScore ↗512+128llamafile 0.9.2Details →
Llama 2 7BQ4_083.1886llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
LLaMA v2Q4_062.5692LocalScore ↗512+128llamafile 0.9.3Details →
Meta Llama 3.1 8B InstructQ4_K - Medium58.8698LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium56.1704LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium55.6672LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium55.5655LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium54.8652LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium54.2663LocalScore ↗512+128llamafile 0.9.2Details →
Llama 2 7BQ8_054.0892llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Meta Llama 3.1 8B InstructQ4_K - Medium53.1627LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium52.4636LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium51.6607LocalScore ↗512+128llamafile 0.9.1Details →
Meta Llama 3.1 8B InstructQ4_K - Medium51.2590LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium49.6583LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium48.6577LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium43.0526LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium41.8478LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium32.2378LocalScore ↗512+128llamafile 0.9.2Details →
Llama 2 7BF1631.6923llama.cpp (Metal) ↗512+128llama.cpp MetalDetails →
Qwen2.5 14B InstructQ4_K - Medium31.3378LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium31.2378LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium31.1360LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium31.1359LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium31.0351LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium30.8343LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium30.8368LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium30.8338LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium30.5349LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium30.4310LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium29.6339LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium29.3324LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium29.1321LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium28.5302LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium28.5309LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium28.3308LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium28.1304LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium28.0328LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium26.6301LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium26.5299LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium26.4299LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium24.8268LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium23.6257LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium23.1255LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium18.6208LocalScore ↗512+128llamafile 0.9.2Details →

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.