Information Extraction and Aggregation on JSON production APIs and synthetic reviews
84.8UtilizationGPT-5.2
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-5.2Defense=Baseline2026.07 | 84.8 | 41.8 | — | — | — | |
| Gemini 3 ProDefense=Baseline2026.07 | 84.8 | 31.3 | — | — | — | |
| GPT-5.2Defense=Randomization2026.07 | 83.9 | — | 3 | 1.5 | 1.5 | |
| GPT-5-miniDefense=Randomization2026.07 | 83.9 | — | 0 | 0 | 0 | |
| Claude Sonnet 4.5Defense=Baseline2026.07 | 83.9 | 37.3 | — | — | — | |
| Gemini 3 ProDefense=Randomization2026.07 | 83.9 | — | 0 | 3 | 1.5 | |
| Gemini 3 FlashDefense=Baseline2026.07 | 83.9 | 43.3 | — | — | — | |
| Gemini 3 FlashDefense=Randomization2026.07 | 83.9 | — | 0 | 3 | 0 | |
| GPT-5-miniDefense=Baseline2026.07 | 83 | 40.3 | — | — | — | |
| Claude Opus 4.5Defense=Baseline2026.07 | 81.2 | 34.3 | — | — | — | |
| Claude Sonnet 4.5Defense=Randomization2026.07 | 78.6 | — | 3 | 1.5 | 0 | |
| Claude Opus 4.5Defense=Randomization2026.07 | 74.1 | — | 0 | 0 | 0 | |
| Gemini 3 FlashDefense=Sanitization2026.07 | 72.3 | 3 | — | — | — | |
| Claude Opus 4.5Defense=Sanitization2026.07 | 71.4 | 0 | — | — | — | |
| GPT-5-miniDefense=Sanitization2026.07 | 70.5 | 0 | — | — | — | |
| Claude Sonnet 4.5Defense=Sanitization2026.07 | 70.5 | 1.5 | — | — | — | |
| Gemini 3 ProDefense=Sanitization2026.07 | 69.6 | 1.5 | — | — | — | |
| GPT-5.2Defense=Sanitization2026.07 | 67.9 | 3 | — | — | — |