Hardware /

RTX 3090

113measured results
377best gen tok/s*
24 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 - Medium37715,607LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium36815,252LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium36715,400LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium36114,974LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium36015,064LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium35014,749LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium34814,452LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium34213,768LocalScore ↗4,352llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium34214,411LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium34013,805LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium33914,601LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium33113,990LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium33014,061LocalScore ↗512+128llamafile 0.9.1Details →
Llama 3.2 1B InstructQ4_K - Medium32814,169LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium32113,357LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium31913,125LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium30812,344LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium30112,964LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ6_K30013,357LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 0.5B InstructQ6_K30019,958LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium27612,950LocalScore ↗512+128llamafile 0.9.3Details →
Llama 3.2 1B InstructQ4_K - Medium21612,560LocalScore ↗4,352llamafile 0.9.3Details →
Llama 3.2 1B InstructQ4_K - Medium20812,930LocalScore ↗4,352llamafile 0.9.3Details →
Qwen2.5 1.5B InstructQ6_K20711,291LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 1.5B InstructQ6_K20511,146LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 1.5B InstructQ6_K20411,246LocalScore ↗512+128llamafile 0.9.2Details →
Gemma 3 1b ItQ4_K - Medium19917,009LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium19413,974LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium19314,107LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium18613,196LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium17612,960LocalScore ↗512+128llamafile 0.9.3Details →
Llama 2 7BQ4_01725,201llama.cpp (Vulkan) ↗512+128llama.cpp Vulkan @d05fe1d — coopmat2Details →
Llama 2 7BQ4_01644,666llama.cpp (Vulkan) ↗512+128llama.cpp Vulkan @d05fe1d — coopmat2Details →
Llama 2 7BQ4_01625,560llama.cpp (CUDA) ↗512+128llama.cpp CUDA @c76b420Details →
Llama 2 7BQ4_01585,175llama.cpp (CUDA) ↗512+128llama.cpp CUDA @c76b420Details →
Llama 3.2 1B InstructQ4_K - Medium15111,382LocalScore ↗512+128llamafile 0.9.3Details →
Phi 4 Mini InstructQ4_K - Medium1455,535LocalScore ↗512+128llamafile 0.9.2Details →
Phi 3 Mini 128k InstructQ6_K1324,481LocalScore ↗512+128llamafile 0.9.2Details →
Phi 3.5 Mini InstructQ6_K1314,403LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1173,895LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1173,926LocalScore ↗4,352llamafile 0.9.2Details →
Qwen3-4BQ4_K - Medium1174,467LocalScore ↗512+128llamafile 0.9.3Details →
Llama 3.2 1B InstructQ4_K - Medium11511,478LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium11411,568LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1133,991LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium11212,246LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1103,895LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1103,898LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1093,858LocalScore ↗512+128llamafile 0.9.2Details →
Llama 3.2 1B InstructQ4_K - Medium10511,674LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1033,552LocalScore ↗512+128llamafile 0.9.1Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1033,637LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1033,511LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1013,454LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium1013,307LocalScore ↗4,352llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium85.43,876LocalScore ↗512+128llamafile 0.9.2Details →
EuroLLM 9B InstructQ4_K - Medium83.92,827LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium83.13,641LocalScore ↗512+128llamafile 0.9.2Details →
Meta Llama 3.1 8B InstructQ4_K - Medium79.73,363LocalScore ↗4,352llamafile 0.9.3Details →
Meta Llama 3.1 8B Instruct AbliteratedQ4_K - Medium77.62,918LocalScore ↗512+128llamafile 0.9.2Details →
Granite 3.2 8b InstructQ6_K73.72,744LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium68.12,187LocalScore ↗4,352llamafile 0.9.2Details →
Qwen3 4BQ4_K - Medium66.13,933LocalScore ↗512+128llamafile 0.9.3Details →
Meta Llama 3.1 8B InstructQ4_K - Medium65.42,251LocalScore ↗512+128llamafile 0.9.2Details →
Qwen3 4BQ4_K - Medium65.24,397LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium65.12,185LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium63.72,135LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium63.62,151LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium63.42,131LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium62.02,126LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium61.92,003LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium61.82,106LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium61.62,083LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium61.01,917LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium60.61,911LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium60.52,034LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.92,005LocalScore ↗512+128llamafile 0.9.1Details →
Qwen2.5 14B InstructQ4_K - Medium59.82,030LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.82,010LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.71,837LocalScore ↗512+128llamafile 0.9.2Details →
Qwen3 8BQ4_K - Medium59.72,970LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium59.62,049LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.41,944LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.11,876LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium59.02,007LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium58.61,878LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium58.41,873LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium58.41,938LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium58.31,854LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium58.21,821LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium56.71,877LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium55.51,884LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium55.21,728LocalScore ↗512+128llamafile 0.9.3Details →
Gemma IfableQ4_K - Medium54.72,318LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium53.21,882LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium52.31,719LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium52.11,705LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium50.72,189LocalScore ↗512+128llamafile 0.9.2Details →
AReaL Boba RL 7BF1650.23,754LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium48.71,957LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium48.71,943LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium48.11,916LocalScore ↗4,352llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium48.01,923LocalScore ↗4,352llamafile 0.9.3Details →
Llama 3.2 1B InstructQ4_K - Medium45.38,635LocalScore ↗4,352llamafile 0.9.3Details →
Qwen3 30B A3BQ4_K - Medium43.2814LocalScore ↗512+128llamafile 0.9.3Details →
Qwen2.5 14B InstructQ4_K - Medium40.91,852LocalScore ↗512+128llamafile 0.9.3Details →
Qwen3 14BQ3_K - Large39.31,604LocalScore ↗512+128llamafile 0.9.3Details →
QwQ 32BQ4_K - Medium30.6915LocalScore ↗4,352llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium27.11,692LocalScore ↗512+128llamafile 0.9.2Details →
Qwen3-30B-A3B-Instruct-2507Q4_K - Medium27.0579LocalScore ↗512+128llamafile 0.9.3Details →
Meta Llama 3.1 8B InstructQ4_K - Medium25.12,540LocalScore ↗4,352llamafile 0.9.3Details →
Fallen Gemma3 27B v1Q4_K - Medium25.1835LocalScore ↗512+128llamafile 0.9.2Details →
Qwen2.5 14B InstructQ4_K - Medium13.91,383LocalScore ↗4,352llamafile 0.9.3Details →

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.