ALL TEN BENCHMARK SETTINGS
Agents & budget leaderboards
See how agent rankings change when only a selected part of a benchmark is evaluated.
Loading recorded data…
Ranked by prefix success rate, highest first. Full-suite success rate and rank are shown alongside it. This is a retrospective ranking of recorded agent runs—not a current live leaderboard. Equal scores share a rank; prefix selection can bias rankings. Historical Mid-Range uses a different leave-target-out subset for each agent, so those rows do not share an identical task set.
| Budget rank | Agent / system | Prefix success rate | Full-suite success rate | Full rank | Rank change | Failures found | Absolute error (pp) |
|---|