Reward Prediction on UI-TARS-7B trajectories
96.5AccuracyPRORE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PRORELast N States=Proactive2025.09 | 96.5 | 66.7 | |
| WebRLLast N States=Full2025.09 | 93.9 | 22.2 | |
| WebRLLast N States=12025.09 | 93 | 20 | |
| Step-CriticLast N States=12025.09 | 93 | 20 | |
| Step-CriticLast N States=Full2025.09 | 93 | 20 | |
| DistRLLast N States=Full2025.09 | 89.5 | 14.3 | |
| DigiRLLast N States=Full2025.09 | 86.8 | 11.7 | |
| DistRLLast N States=12025.09 | 84.2 | 25 | |
| DigiRLLast N States=12025.09 | 84.2 | 10 |