Long-term Planning on GUI-Odyssey
66.26Type Success RateRL-Continuous-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RL-Continuous-7Bmode=Post-Training, reward=Continuous, paradigm=RLVR2026.01 | 66.26 | 43.33 | 38.61 | 55.09 | |
| SSL-7Bmode=Post-Training, reward=Sweet Spot Learning (SSL), paradigm=RLVR2026.01 | 65.9 | 46.02 | 39.93 | 56.92 | |
| SSL-3Bmode=Post-Training, reward=Sweet Spot Learning (SSL), paradigm=RLVR2026.01 | 63.75 | 46.44 | 42.47 | 57.11 | |
| RL-Binary-7Bmode=Post-Training, reward=Binary, paradigm=RLVR2026.01 | 63.47 | 44.79 | 37.67 | 54.95 | |
| OS-Atlas-7Bmode=Post-Training2026.01 | 60.42 | 39.74 | 26.96 | 44.88 | |
| RL-Continuous-3Bmode=Post-Training, reward=Continuous, paradigm=RLVR2026.01 | 58.57 | 43.97 | 36.69 | 52.47 | |
| QwenVL2.5-7B*mode=Post-Training, SFT on Mix-3K=true2026.01 | 56.78 | 38.65 | 34.44 | 50.97 | |
| RL-Binary-3Bmode=Post-Training, reward=Binary, paradigm=RLVR2026.01 | 56.78 | 42.67 | 38.78 | 49.81 | |
| QwenVL2.5-7Bmode=Zero Shot2026.01 | 55.6 | 37.78 | 34.37 | 50.53 | |
| UI-R1-3Bmode=Post-Training2026.01 | 52.16 | 34.46 | 32.49 | 46.35 | |
| OS-Atlas-4Bmode=Post-Training2026.01 | 49.63 | 34.63 | 20.25 | 37.63 | |
| QwenVL2.5-3B*mode=Post-Training, SFT on Mix-3K=true2026.01 | 43.69 | 32.21 | 27.31 | 41 | |
| GPT-4omode=Zero Shot2026.01 | 37.5 | 14.17 | 5.36 | 28.69 | |
| QwenVL2.5-3Bmode=Zero Shot2026.01 | 37.4 | 26.49 | 26.69 | 37.3 |