Web Task Success on Mind2Web Live
34.7Task Success RatePRO-CUA
Evaluation Results
| Method | Links | |
|---|---|---|
| PRO-CUATraining Paradigm=Self-evolving 4B, Ext. Expert Steps=02026.05 | 34.7 | |
| PRO-CUA-8Training Paradigm=Self-evolving 8B, Ext. Expert Steps=02026.05 | 30.6 | |
| Rule-based Step-RLTraining Paradigm=Self-evolving 4B, Ext. Expert Steps=02026.05 | 27.8 | |
| FBCTraining Paradigm=Self-evolving 4B, Ext. Expert Steps=02026.05 | 26.4 | |
| Rule-based Step-RLTraining Paradigm=Self-evolving 8B, Ext. Expert Steps=02026.05 | 25 | |
| FBCTraining Paradigm=Self-evolving 8B, Ext. Expert Steps=02026.05 | 23.6 | |
| Qwen3-VL-8B-InstructTraining Paradigm=Self-evolving 8B, Ext. Expert Steps=02026.05 | 20.8 | |
| WebSTAR-32BTraining Paradigm=External expert / closed data, Ext. Expert Steps=100K2026.05 | 20.4 | |
| UI-TARS-1.5-7BTraining Paradigm=External expert / closed data, Ext. Expert Steps=Closed data2026.05 | 18.1 | |
| Qwen3-VL-4B-InstructTraining Paradigm=Self-evolving 4B, Ext. Expert Steps=02026.05 | 18.1 | |
| WebSTAR-7BTraining Paradigm=External expert / closed data, Ext. Expert Steps=100K2026.05 | 17 |