Reward Modeling on OSWorld Verified Class-Imbalanced Test Scripts 1.0 (test)
61.9PrecisionVAGEN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VAGENVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 61.9 | 86.4 | 72.2 | 85 | |
| ZeroGUIVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 52.3 | 71.6 | 60.4 | 78.9 | |
| FullTrajEvalVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 50.5 | 69.1 | 58.3 | 77.8 | |
| AndroidGenVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 45.3 | 48.1 | 46.7 | 75.3 | |
| WebRLVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 44.2 | 42 | 43 | 75.1 | |
| DistRLVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 43.4 | 40.7 | 42 | 74.8 | |
| DigiRLVerified Agent Backbone=UI-TARS-1.5-7B2026.01 | 39.2 | 38.3 | 38.7 | 72.9 |