GUI Navigation on Mind2Web Online (Average)
64Success RateOvis2.5S-GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ovis2.5S-GRPOTraining Method=GRPO (Step-wise)2026.02 | 64 | 11.33 | |
| Ovis2.5S-RF++Training Method=REINFORCE++ (Step-wise)2026.02 | 64 | 11.33 | |
| Ovis2.5S-RLOOTraining Method=RLOO (Step-wise)2026.02 | 62.67 | 10 | |
| Claude 4 Sonnet CUModel Type=Proprietary2026.02 | 62.33 | — | |
| Claude 3.7 Sonnet CUModel Type=Proprietary2026.02 | 61 | — | |
| Ovis2.5SFTTraining Method=SFT2026.02 | 52.67 | — | |
| GPT-4oModel Type=Proprietary2026.02 | 37 | — | |
| PageGuideBackbone=google/gemini-3-flash-preview2026.04 | 35.17 | — | |
| UI-TARSModel Scale=7B, Version=1.52026.02 | 33.33 | — | |
| PageGuideBackbone=google/gemini-2.5-flash-preview2026.04 | 30.42 | — | |
| SeeActBackbone=GPT-42026.04 | 30 | — | |
| Qwen3-VLModel Scale=32B2026.02 | 27.67 | — | |
| Qwen3-VLModel Scale=8B2026.02 | 24.33 | — |