Local AI Benchmarks

Loading…

Runs
–
benchmark runs in view
Models
–
distinct models
Best prose speed
– tps
–
Best code speed
– tps
–

Fastest machine

Best prose decode speed per machine (tokens/s)

Prose vs code

Best run per machine (tokens/s) prosecode

Every run

All runs by recipe, sorted by prose speed (tokens/s)

Prose vs code per run

Each dot is one run; dashed line marks code = prose 16 GB32 GB128 GB256 GB

Models

Best prose speed per model (tokens/s)

All runs

One row per run — click a column header to sort

Class Machine Model Recipe Prose (tps) Code (tps)