GUI Agent Task Execution on Live mini-programs
49Success RateGemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashGrounding=w/ Grounding2026.02 | 49 | |
| Claude-4.5-OpusGrounding=w/ Grounding2026.02 | 46.3 | |
| UI-OceanusModel Size=32B, Stage=Multi-step GRPO2026.02 | 44.3 | |
| UI-OceanusModel Size=32B, Stage=Step-level GRPO2026.02 | 38.9 | |
| UI-OceanusModel Size=32B, Stage=SFT Cold Start2026.02 | 29.5 | |
| Seed-1.8Grounding=w/ Grounding2026.02 | 27.5 |