GUI reasoning on Aggregate (GUI-Act-Web, OmniAct-Web, AndroidControl)
74.6Overall AccuracyCAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| CAPOPrompting=zero-shot2025.12 | 74.6 | |
| GRPOPrompting=zero-shot2025.12 | 70.79 | |
| QwenVL2.5-3BPrompting=zero-shot2025.12 | 54.09 | |
| Os-Atlas-4BPrompting=zero-shot2025.12 | 49.75 |