GUI Agent Interaction on OSWorld w/ Loop
6.9AUVClaude3.7-Sonnet
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude3.7-Sonnet2026.02 | 6.9 | 31.1 | |
| GPT-4o + ScaleCUA-7BPlanner=GPT-4o, GUI grounder=ScaleCUA-7B2026.02 | 6.6 | 47.4 | |
| UI-TARS-72B-DPO2026.02 | 5.3 | 50.7 | |
| UI-TARS-1.5-7B2026.02 | 3.2 | 57.2 | |
| Qwen2.5-VL-72B-Instruct2026.02 | 0.4 | 50.4 |