GUI Navigation on Online-Mind2Web (Hard)
53.52Success RateClaude 4 Sonnet CU
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude 4 Sonnet CUModel Type=Proprietary2026.02 | 53.52 | |
| Ovis2.5S-GRPOTraining Method=GRPO (Step-wise)2026.02 | 51.35 | |
| Ovis2.5S-RLOOTraining Method=RLOO (Step-wise)2026.02 | 51.35 | |
| Claude 3.7 Sonnet CUModel Type=Proprietary2026.02 | 46.48 | |
| Ovis2.5S-RF++Training Method=REINFORCE++ (Step-wise)2026.02 | 44.59 | |
| Ovis2.5SFTTraining Method=SFT2026.02 | 39.19 | |
| UI-TARSModel Scale=7B, Version=1.52026.02 | 17.57 | |
| Qwen3-VLModel Scale=32B2026.02 | 14.86 | |
| Qwen3-VLModel Scale=8B2026.02 | 9.46 |