GUI Element Localization on ScreenSpot v2
92.8AccuracyKimi-VL-16B-MoE
Evaluation Results
| Method | Links | |
|---|---|---|
| Kimi-VL-16B-MoEBackbone=16B-MoE2025.05 | 92.8 | |
| ViGoRL-7bBackbone=7b2025.05 | 91 | |
| ViGoRL-3bBackbone=3b2025.05 | 86.5 | |
| ViGoRL-3bBackbone=3b, Feedback Loop=Multi-turn RL2025.05 | 86.1 | |
| UI-TARS-2BBackbone=2B2025.05 | 84.7 | |
| Qwen2.5-VL-3B + Vanilla GRPOBackbone=3B, Training Protocol=Vanilla GRPO2025.05 | 84.4 | |
| OS-Atlas-7BBackbone=7B2025.05 | 84.1 | |
| Qwen2.5-VL-3B + SFT directBackbone=3B, Training Protocol=SFT direct2025.05 | 80.6 | |
| Qwen2.5-VL-7BBackbone=7B2025.05 | 73.6 | |
| OS-Atlas-4BBackbone=4B2025.05 | 71.9 | |
| Qwen2.5-VL-3BBackbone=3B2025.05 | 68.4 | |
| SeeClick2025.05 | 55.1 | |
| GPT-4o2025.05 | 18.1 |