Attack Success Rate (ASR_B) on VisualWebArena Task A pseudo trajectories
19.5Action RateGPT-OSS-120B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-OSS-120B2026.04 | 19.5 | 14.5 | 6 | 14.2 | 8.2 | |
| GPT-5-mini2026.04 | 4.6 | 2.5 | 3.6 | 32.5 | 28.9 | |
| Qwen3-32B2026.04 | 4.3 | 5.7 | 3.2 | 3.9 | 0.7 | |
| GPT-5.22026.04 | 1.8 | 22.3 | 6.4 | 23.4 | 17 | |
| Qwen3.5-122B2026.04 | 1.8 | 0 | 2.1 | 12 | 9.9 | |
| Qwen2.5-72B2026.04 | 0 | 0 | 0 | 0.7 | 0.7 |