Information Extraction and Aggregation on Web DOM Nanobrowser based synthesized element format
100UtilityClaude Opus 4.5
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Claude Opus 4.5Defense=Baseline2026.07 | 100 | 33.3 | — | — | — | |
| Claude Opus 4.5Defense=Randomization2026.07 | 100 | — | 0 | 73.3 | 0 | |
| Claude Sonnet 4.5Defense=Baseline2026.07 | 100 | 60 | — | — | — | |
| Claude Sonnet 4.5Defense=Randomization2026.07 | 100 | — | 0 | 93.3 | 0 | |
| GPT-5.2Defense=Baseline2026.07 | 97.8 | 100 | — | — | — | |
| GPT-5.2Defense=Randomization2026.07 | 97.8 | — | 0 | 100 | 0 | |
| GPT-5-miniDefense=Baseline2026.07 | 97.8 | 100 | — | — | — | |
| GPT-5-miniDefense=Randomization2026.07 | 97.8 | — | 0 | 100 | 0 | |
| Gemini 3 ProDefense=Baseline2026.07 | 97.8 | 33.3 | — | — | — | |
| Gemini 3 ProDefense=Randomization2026.07 | 97.8 | — | 0 | 60 | 0 | |
| Gemini 3 FlashDefense=Baseline2026.07 | 97.8 | 93.3 | — | — | — | |
| Gemini 3 FlashDefense=Randomization2026.07 | 97.8 | — | 0 | 100 | 0 | |
| Gemini 3 FlashDefense=Sanitization2026.07 | 80.6 | 18.3 | — | — | — | |
| Claude Sonnet 4.5Defense=Sanitization2026.07 | 77.8 | 26.7 | — | — | — | |
| Claude Opus 4.5Defense=Sanitization2026.07 | 75 | 0 | — | — | — | |
| GPT-5-miniDefense=Sanitization2026.07 | 73.3 | 0 | — | — | — | |
| GPT-5.2Defense=Sanitization2026.07 | 68.3 | 8.3 | — | — | — | |
| Gemini 3 ProDefense=Sanitization2026.07 | 67.2 | 0 | — | — | — |