Global Rankings

Math rankings

568 models · sorted by Elo · click any column to re-sort

Showing math benchmark columns only.

Elo = Bradley-Terry rating over pairwise benchmark battles: every benchmark two models both report is a head-to-head, weighted so no domain decides a pairing linearly by benchmark count (a shared domain votes with √n total weight), and 400 points ≈ 10:1 expected win odds. Models with too few results to compare fairly are shown unrated and sorted last.