Agreement with outcome human labels on CUAVerifierBench Internal Dataset
81AccuracyUniversal Verifier (UV)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Universal Verifier (UV)Verifier=UV, Backbone=GPT-5.22026.04 | 81 | 81 | 64 | 32 | 1 | |
| Universal VerifierVerifier=UV, Base Model=GPT-5.22026.04 | 81 | 81 | 64 | 32 | 1 | |
| WebJudgeVerifier=WebJudge, Backbone=o4-mini2026.04 | 72 | 74 | 44 | 33 | 22 | |
| WebJudgeVerifier=WebJudge, Base Model=o4-mini2026.04 | 72 | 74 | 44 | 33 | 22 | |
| WebVoy.Verifier=WebVoy., Backbone=GPT-5.22026.04 | 70 | 69 | 43 | 44 | 10 | |
| WebVoy.Verifier=WebVoy., Backbone=GPT-4o2026.04 | 67 | 73 | 31 | 24 | 45 | |
| WebVoyagerVerifier=WebVoy., Base Model=GPT-4o2026.04 | 67 | 73 | 31 | 24 | 45 | |
| WebJudgeVerifier=WebJudge, Backbone=GPT-5.22026.04 | 64 | 58 | 33 | 57 | 7 |