GUI planning semantic consistency on GUIGuard-Bench Overall 1.0 (Mixed)
1.16Black Mask ScoreUI-Tars-1.5-7B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UI-Tars-1.5-7BCategory=GUI Agent2026.01 | 1.16 | 1.44 | 0.88 | 1.01 | 1.11 | |
| Fara-7BCategory=GUI Agent2026.01 | 2.02 | 1.94 | 2.11 | 1.95 | 2.04 | |
| GUI-Owl-7BCategory=GUI Agent2026.01 | 2.56 | 2.5 | 2.85 | 2.32 | 2.55 | |
| Qwen3-VL-235B-A22BCategory=Open-source VLM2026.01 | 2.71 | 2.92 | 2.63 | 2.63 | 2.73 | |
| Claude-Sonnet-4.5Category=Closed-source VLM2026.01 | 2.87 | 2.74 | 3 | 3.03 | 2.96 | |
| Gemini-3-ProCategory=Closed-source VLM2026.01 | 3.32 | 3.42 | 3.4 | 3.22 | 3.35 | |
| GPT-5.2Category=Closed-source VLM2026.01 | 3.33 | 3.3 | 3.44 | 3.17 | 3.31 |