Submit a run

Every speed on this site is currently an estimate. A measured run from your Mac is worth more than any model we can write. Fill this in and copy the result — we publish it as unverified until three independent runs agree.

ChipUnified memoryModelQuantizationRuntimeRuntime versionGeneration speed (tok/s)Prompt speed (tok/s)Context used (tokens)

Anything unusual? (thermals, other apps, power)

Your submission

chip: M1 (m1)
memory: 8 GB
model: Qwen2.5 0.5B Instruct (qwen2.5-0.5b-instruct)
quant: q4
runtime: mlx-lm 
context: 8192 tokens
decode:  tok/s
prefill:  tok/s

Post it publicly

Read how a run gets verified before timing: warm start, at least 256 generated tokens, median of three runs, mains power.