Loading page content
AutomationBench (Private)Agents
Uses the private held-out evaluation set from the official AutomationBench leaderboard. It is not the same dataset as the 600 public repository tasks, and private tasks may change between versions, so these scores are not combined with public-set results.AutomationBench (Private) is the private held-out evaluation set used by the official AutomationBench leaderboard. Its task distribution resembles the public 600-task set, but it is intentionally harder and not released. A task passes only when all assertions succeed.
#ModelScore
4 models total