Leaderboards are useful for a shortlist but disagree with each other, and newer models can be missing or scored under different settings. This site shows the source of every score it uses.
Frequently asked questions
What does Leaderboard mean?
A public ranking of models on one or more benchmarks. Leaderboards are useful for a shortlist but disagree with each other, and newer models can be missing or scored under different settings. This site shows the source of every score it uses.