Pairs are generated between frontier models from different organizations released within about two years of each other — the comparisons that are actually meaningful. Each page puts their specifications side by side and lists every benchmark both were scored on.
