Leaderboard

Bucket: agent × model × effort × sandbox. Cross-sandbox scores are not comparable.