GUI Element Localization on ScreenSpot Pro
34.5AccuracyKimi-VL-16B-MoE
Evaluation Results
| Method | Links | |
|---|---|---|
| Kimi-VL-16B-MoEBackbone=16B-MoE2025.05 | 34.5 | |
| ViGoRL-7bBackbone=7b2025.05 | 33.1 | |
| ViGoRL-3bBackbone=3b, Feedback Loop=Multi-turn RL2025.05 | 32.3 | |
| UGround-V1-7BBackbone=7B2025.05 | 31.1 | |
| ViGoRL-3bBackbone=3b2025.05 | 31.1 | |
| Qwen2.5-VL-3B + Vanilla GRPOBackbone=3B, Training Protocol=Vanilla GRPO2025.05 | 29 | |
| Qwen2.5-VL-7BBackbone=7B2025.05 | 29 | |
| UI-TARS-2BBackbone=2B2025.05 | 27.7 | |
| Qwen2.5-VL-3B + SFT directBackbone=3B, Training Protocol=SFT direct2025.05 | 25 | |
| Qwen2.5-VL-3BBackbone=3B2025.05 | 23.9 | |
| OS-Atlas-7BBackbone=7B2025.05 | 18.9 | |
| UI-R1-3BBackbone=3B2025.05 | 17.8 | |
| Claude Comp. Use2025.05 | 17.1 | |
| UGround-7BBackbone=7B2025.05 | 16.5 | |
| CogAgent-18BBackbone=18B2025.05 | 7.7 | |
| ShowUI-2BBackbone=2B2025.05 | 7.7 | |
| OS-Atlas-4BBackbone=4B2025.05 | 3.7 | |
| Qwen2-VL-7BBackbone=7B2025.05 | 1.6 | |
| SeeClick2025.05 | 1.1 | |
| GPT-4o2025.05 | 0.8 |