Two ways to win
The benchmark is run like a tournament with two medal tracks at once — every model ranked on its own, and lab teams in a knockout bracket — because companies adopt a vendor's lineup, not one model.
One page for every reader — the role changes what leads, the depth changes how much shows. Nothing is held back: every section can show the rest in place.
Two tracks, one run
Individual: all 33 models are ranked on their own work. Team: 8 labs each field a three-model squad that competes in a knockout bracket.
Who is in the bracket
The bracket's 8 teams are the labs with a full three-tier lineup: Anthropic, OpenAI, Google, DeepSeek, Qwen, Moonshot, Z.ai and MiniMax.
How the bracket runs
A classic knockout: quarterfinals (8 → 4), semifinals (4 → 2), final (2 → 1) — one champion, no byes.
Breaking ties
This section goes deeper than the current view.
Where it stands
No round has been scored, so there are no standings, no winners and no champion yet. This page describes the format only.