Desktop Application GUI Agent on Chrome 45 tasks
77.8Pass RateTrajectory Distillation + Experience Bank
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Trajectory Distillation + Experience BankBackbone=Qwen3-VL-8B, Iterations=12026.04 | 77.8 | 0.77 | 0.724 | 36.3 | |
| Experience BankBackbone=Qwen3-VL-8B, Iterations=1, Mode=inference-time only2026.04 | 62.2 | 0.64 | 0.646 | 45 | |
| BaselineBackbone=Qwen3-VL-8B, Iterations=12026.04 | 60 | 0.654 | 0.662 | 43.7 |