Step Accuracy on AndroidControl High Level v2
64.3Pass@1GUI-Libra-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GUI-Libra-8BType=Ours2026.02 | 64.3 | 70.6 | |
| GUI-Libra-4BType=Ours2026.02 | 62.3 | 68.6 | |
| GPT-5 + UGround-v1-7BFramework=SeeAct-V, Type=Proprietary2026.02 | 61.3 | 69.4 | |
| GUI-Libra-7BType=Ours2026.02 | 59.3 | 67.3 | |
| Qwen3-VL-32BType=Open-source Native2026.02 | 58.8 | 69.6 | |
| GPT-4.1 + UGround-v1-7BFramework=SeeAct-V, Type=Proprietary2026.02 | 57.5 | 63.3 | |
| GUI-Libra-3BType=Ours2026.02 | 57.3 | 67.1 | |
| GPT-4o + UGround-v1-7BFramework=SeeAct-V, Type=Proprietary2026.02 | 57 | 66.3 | |
| Qwen2.5-VL-72BType=Open-source Native2026.02 | 56.5 | 72.9 | |
| Qwen3-VL-8BType=Baseline2026.02 | 54.8 | 66.1 | |
| GPT-5-mini + UGround-v1-7BFramework=SeeAct-V, Type=Proprietary2026.02 | 52.8 | 58.8 | |
| Qwen3-VL-4BType=Baseline2026.02 | 49.3 | 63.3 | |
| Qwen2.5-VL-32BType=Open-source Native2026.02 | 49 | 66.6 | |
| Qwen2.5-VL-7BType=Baseline2026.02 | 46.5 | 58.5 | |
| UI-TARS-1.5-7BType=Open-source Native2026.02 | 45.2 | 63.1 | |
| GUI-R1-3BType=Open-source Native2026.02 | 40 | 54 | |
| GUI-R1-7BType=Open-source Native2026.02 | 39.7 | 56.3 | |
| Aguvis-7BType=Open-source Native2026.02 | 37.7 | 43.7 | |
| GLM-4.1V-9B-ThinkingType=Open-source Native2026.02 | 37.2 | 49 | |
| Qwen2.5-VL-3BType=Baseline2026.02 | 36.4 | 50.8 |