Online Reinforcement Learning on AndroidWorld
62.28Success RateStainFlow
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| StainFlowVerifier=Qwen3.5-27B2026.06 | 62.28 | 0.81 | 0.39 | 0.42 | 18.36 | |
| StainFlowVerifier=Qwen3.5-9B2026.06 | 60.34 | 0.78 | 0.41 | 0.37 | 18.62 | |
| OS-ThemisVerifier=Qwen3.5-27B2026.06 | 60.34 | 0.38 | 0.29 | 0.09 | 18.05 | |
| OS-ThemisVerifier=Qwen3.5-9B2026.06 | 58.62 | 0.36 | 0.22 | 0.14 | 18.56 | |
| ADMIREVerifier=Qwen3.5-27B2026.06 | 57.76 | 0.37 | 0.15 | 0.22 | 20.21 | |
| ADMIREVerifier=Qwen3.5-9B2026.06 | 56.9 | 0.29 | 0.13 | 0.16 | 20.45 | |
| GUI-Critic-R1Verifier=GUI-Critic-R12026.06 | 56.03 | 0.61 | 0.33 | 0.28 | 20.52 | |
| Outcome RewardVerifier=Environment2026.06 | 54.31 | — | — | — | 20.95 | |
| No TrainingVerifier=N/A2026.06 | 48.28 | — | — | — | 21.73 |