GUI Agent Action Execution on ScreenDrag 1.0 (online)
99.01Success Rate (OS)OpenCUA-7B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| OpenCUA-7BAction Paradigm=Action as Language, Model Type=Open-source models2025.12 | 99.01 | 4.95 | 0 | 5.94 | 0 | 21.98 | |
| OpenCUA-32BAction Paradigm=Action as Language, Model Type=Open-source models2025.12 | 97.03 | 6.93 | 0 | 0 | 0 | 20.79 | |
| Gemini-2.5-CUAAction Paradigm=Action as Language, Model Type=Proprietary models2025.12 | 86.14 | 20.79 | 0 | 3.96 | 0 | 22.18 | |
| Qwen3-VL-32BAction Paradigm=Action as Language, Model Type=Open-source models2025.12 | 83.17 | 4.95 | 3.96 | 2.97 | 0 | 19.01 | |
| Seed-1.6-VisionAction Paradigm=Action as Language, Model Type=Proprietary models2025.12 | 77.23 | 3.96 | 1.98 | 8.91 | 2.97 | 19.01 | |
| UI-TARS-1.5-7BAction Paradigm=Action as Language, Model Type=Proprietary models2025.12 | 73.27 | 1.98 | 1.98 | 7.92 | 0 | 17.03 | |
| OperatorAction Paradigm=Action as Language, Model Type=Proprietary models2025.12 | 53.47 | 9.9 | 2.97 | 0 | 0 | 13.27 | |
| Qwen3-VL-8BAction Paradigm=Action as Language, Model Type=Open-source models2025.12 | 24.75 | 5.94 | 6.93 | 0 | 0 | 7.52 | |
| Qwen3-VL-2BAction Paradigm=Action as Language, Model Type=Open-source models2025.12 | 13.86 | 6.93 | 2.97 | 0 | 0 | 4.75 | |
| ShowUI-π-450MAction Paradigm=Flow-based action, Model Type=Open-source models2025.12 | 13.11 | 22.93 | 8.64 | 55.91 | 34.32 | 26.98 |