BenchmarkAgentRewardBench
AgentRewardBench
A benchmark for evaluating the ability to judge task completion by web agents, containing expert-annotated agent trajectory data.
Timeline (last 90 days)
Oct 17
On the 1,302 expert-annotated trajectories of AgentRewardBench, JEV-27B-VL reports an AUROC of 0.91, with precision 78.4 and recall 70.2 at a threshold of 0.5.
Unverified50%