GUI Grounding on Mind2Web (M2W)
16Convolutional StepsLLaDA-V (Lin)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaDA-V (Lin)Parameter Count=8B, Masking Strategy=Linear2026.03 | 16 | 82.4 | 98.5 | 3.02 | |
| PhiParameter Count=3B2026.03 | 21 | 56.8 | 94.4 | — | |
| Qwen 2.5 VLParameter Count=3B2026.03 | 21 | 79.3 | 99.6 | — | |
| Qwen 2.5 VLParameter Count=7B2026.03 | 21 | 81.9 | 99.9 | 1.1 | |
| LLaDA-V (Ours)Parameter Count=8B, Masking Strategy=Hybrid (Linear and Full Deterministic)2026.03 | 23 | 83.9 | 100 | 5.44 |