UI Comprehension on UI Comprehension-Bench
87.4Locate AccuracyGUI-Owl-7B w/ UILoop
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GUI-Owl-7B w/ UILoopParadigm=Screen-to-Action Training Models2026.04 | 87.4 | 51.1 | 53.4 | 23.8 | |
| UILoop-7BParadigm=UILoop Training Models2026.04 | 86.4 | 49.3 | 61.3 | 26.1 | |
| UILoop-3BParadigm=UILoop Training Models2026.04 | 80.3 | 44.7 | 50.2 | 18 | |
| OS-Atlas-Pro-7B w/ UILoopParadigm=Screen-to-Action Training Models2026.04 | 71.4 | 54.2 | 34.9 | 13.5 | |
| GUI-R1-7BParadigm=Screen-to-Action Training Models2026.04 | 62.6 | 47.6 | 35.3 | 10.5 | |
| GUI-Owl-7BParadigm=Screen-to-Action Training Models2026.04 | 61.9 | 21.1 | 41 | 5.4 | |
| OS-Atlas-Pro-7BParadigm=Screen-to-Action Training Models2026.04 | 49.6 | 48.2 | 18.9 | 4.5 | |
| Qwen2.5-VL-3B-InstructParadigm=Zero-shot Models2026.04 | 48.7 | 9.5 | 36.6 | 1.7 | |
| GUI-R1-3BParadigm=Screen-to-Action Training Models2026.04 | 47.4 | 37.9 | 35.9 | 6.4 | |
| UI-R1-3BParadigm=Screen-to-Action Training Models2026.04 | 47.1 | 39.7 | 33.7 | 6.3 | |
| Qwen2.5-VL-7B-InstructParadigm=Zero-shot Models2026.04 | 46.8 | 27.5 | 29.1 | 3.7 | |
| GPT-4oParadigm=Zero-shot Models2026.04 | 22.5 | 30.7 | 11.8 | 0.8 |