GUI Agent on AndroidControl
85.25TypeLearn from AndroidControl
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Learn from AndroidControlBase Model=OS-Atlas-Pro-7B, Learning Strategy=Supervised Fine-Tuning, Training Source=AndroidControl2026.01 | 85.25 | 57.95 | 20.31 | |
| CL from AITZ and AndroidControlBase Model=OS-Atlas-Pro-7B, Learning Strategy=Continual Learning, Training Source=AITZ, AndroidControl2026.01 | 84.49 | 57.47 | 19.4 | |
| Agent-DiceBase Model=OS-Atlas-Pro-7B, Learning Strategy=Geometric Consensus & Curvature Importance Weighting2026.01 | 80.03 | 51.42 | 14.42 | |
| CL from all threeBase Model=OS-Atlas-Pro-7B, Learning Strategy=Continual Learning, Training Source=AITZ, AndroidControl, GUI-Odyssey2026.01 | 73.56 | 43.48 | 9.57 | |
| Zero-ShotBase Model=OS-Atlas-Pro-7B, Learning Strategy=Zero-Shot2026.01 | 73.12 | 47.14 | 13.63 | |
| Learn from GUI-OdysseyBase Model=OS-Atlas-Pro-7B, Learning Strategy=Supervised Fine-Tuning, Training Source=GUI-Odyssey2026.01 | 69.03 | 36.79 | 6.29 | |
| Learn from AITZBase Model=OS-Atlas-Pro-7B, Learning Strategy=Supervised Fine-Tuning, Training Source=AITZ2026.01 | 61.82 | 36.7 | 7.54 |