Agent Trajectory Verification Agreement on WebVoyager
3.4Unterminated Rate (%)Native Verifier
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Native VerifierAgent Model=GPT-5, N (tasks scored)=5932026.04 | 3.4 | 90.6 | 79.4 | 71 | 4 | 68 | 83 | 90 | 0.36 | 2 | 72 | 78 | 86 | 0.33 | |
| Native VerifierAgent Model=Fara-7B, N (tasks scored)=5942026.04 | 4.2 | 74.6 | 49 | 37.9 | 6 | 56 | 69 | 75 | 0.38 | 1 | 60 | 63 | 68 | 0.33 |