Evidence

What was actually measured.

Pick one or more releases and this view shows the benchmark results recorded for them: the score and its unit, the exact variant and version, when it was evaluated, and who published it. Where two models were measured the same way their results are grouped as comparable; where they were not, the view says so rather than placing numbers on a scale they do not share. There is no overall score and no ranking here.

Which models to explore is read from this page’s address, which needs JavaScript on a site built entirely ahead of time. Every value below is also on each model’s own Model Passport, with the same sources and dates.

No models selected. Choose one or more releases to see their benchmark evidence.

Choose models

Up to 4 models. Each entry is a link, so this works without JavaScript too.

Filter models

Choose a model to see its evidence

This view reads the models to explain from the page address. Pick one or more releases and it will show what each was measured on, with every source and date.

Browse the model catalogue

Start a new evidence view

Every release, with its own page

Which models to explore is read from this page’s address, which needs JavaScript on a site built entirely ahead of time. Every benchmark result is also on each release’s static Model Passport, with the same values, sources, and dates: