Agentic UI Interaction on GUIRILLA-TASK agentic
12.5Input Success RateQwen 2.5 VL 3B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen 2.5 VL 3BModel Size=3B, Base Model=Qwen 2.5 VL2025.10 | 12.5 | 42.95 | 40.77 | |
| Claude Computer Use2025.10 | 8.93 | 65.59 | 61.53 | |
| OpenAI Computer Use2025.10 | 8.04 | 68.75 | 64.41 | |
| OS-Atlas-Pro-7BModel Size=7B2025.10 | 7.14 | 62.84 | 58.85 | |
| UI TARS 2BModel Size=2B2025.10 | 7.14 | 50.24 | 47.16 | |
| CogAgent 9BModel Size=9B2025.10 | 3.57 | 15.83 | 14.95 | |
| Qwen 2.5 VL 7BModel Size=7B, Base Model=Qwen 2.5 VL2025.10 | 2.68 | 39.16 | 36.55 | |
| UI TARS 1.5 7BModel Size=7B, Version=1.52025.10 | 1.79 | 54.65 | 50.86 |