The best language models that fit in 600M parameters.
Nano LM Index ranks small open-weight models on 12 benchmarks across six skills, from knowledge and maths to tool use, code and classification.
Who leads at ≤600M
Models are ranked by their Nano LM Index score, highest first. The small range under each score is its 95% confidence interval; the strongest score in each skill column is outlined. Click a row for its full breakdown.
Where each model is strong
Every cell is one benchmark's chance-corrected skill. Small models are uneven: a model can be strong at maths and still near zero at classification.
Compare up to three models
The radar shows the six category scores; the bars below show every benchmark.
12 benchmarks, 6 skills, equal weight
Every item of every benchmark is used, with no subsampling. Sources and files are pinned by revision and sha256.
One protocol, no judgement calls
Whenever a setting would have to differ between models, the model's own artefacts decide it, never the maintainers. The full standard is published as the protocol.
What's new
Releases, new results and changes to the protocol.