Long-term Planning on AndroidControl High
71.79Type RateSSL-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SSL-7Bmode=Post-Training, reward=Sweet Spot Learning (SSL), paradigm=RLVR2026.01 | 71.79 | 65.56 | 52.31 | 56.92 | |
| RL-Binary-7Bmode=Post-Training, reward=Binary, paradigm=RLVR2026.01 | 70.82 | 63.43 | 49.5 | 54.95 | |
| RL-Continuous-7Bmode=Post-Training, reward=Continuous, paradigm=RLVR2026.01 | 70.45 | 62.77 | 49.09 | 55.09 | |
| QwenVL2.5-7B*mode=Post-Training, SFT on Mix-3K=true2026.01 | 69.15 | 58.69 | 48.11 | 50.97 | |
| QwenVL2.5-7Bmode=Zero Shot2026.01 | 68.67 | 59.71 | 47.06 | 50.53 | |
| SSL-3Bmode=Post-Training, reward=Sweet Spot Learning (SSL), paradigm=RLVR2026.01 | 65.96 | 62.41 | 61.62 | 57.11 | |
| RL-Continuous-3Bmode=Post-Training, reward=Continuous, paradigm=RLVR2026.01 | 64.72 | 60.91 | 49.96 | 52.47 | |
| GPT-4omode=Zero Shot2026.01 | 63.06 | 30.9 | 21.17 | 28.69 | |
| RL-Binary-3Bmode=Post-Training, reward=Binary, paradigm=RLVR2026.01 | 59.56 | 56.24 | 44.85 | 49.81 | |
| UI-R1-3Bmode=Post-Training2026.01 | 57.85 | 55.7 | 45.44 | 46.35 | |
| OS-Atlas-7Bmode=Post-Training2026.01 | 57.44 | 54.9 | 29.83 | 44.88 | |
| QwenVL2.5-3B*mode=Post-Training, SFT on Mix-3K=true2026.01 | 52.05 | 49.53 | 41.22 | 41 | |
| OS-Atlas-4Bmode=Post-Training2026.01 | 49.01 | 49.51 | 22.77 | 37.63 | |
| QwenVL2.5-3Bmode=Zero Shot2026.01 | 47.81 | 46.51 | 38.9 | 37.3 |