List your company

Reward Verification on AgentRewardBench VisualWebArena

100PrecisionRule-based (VWA Oracle)
71.9279.2186.593.79Jul 15, 2025

Evaluation results

MethodLinks
Rule-based (VWA Oracle)version=ours2025.07100
SGVbackbone=GPT-o42025.0786
Rule-based (VWA Oracle)version=original2025.0785
SGVbackbone=Gemini 2.5, mode=no Thinking2025.0780
No-SGV Baselinebackbone=GPT-o42025.0780
WebJudgebackbone=GPT-o4, status=SOTA2025.0775
No-SGV Baselinebackbone=Gemini 2.5, mode=no Thinking2025.0773