Automated auditing on BIXBench (Verified-50)
83.3Recall (A)Ensemble (any)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Ensemble (any)Description=Pooled findings without cross-model deduplication2026.04 | 83.3 | 95.8 | — | — | 14.38 | 383 | |
| Opus 4.62026.04 | 54.2 | 79.2 | 38.7 | 67.7 | 5.98 | 66 | |
| GPT-5.42026.04 | 50 | 87.5 | 23.3 | 55.8 | 1.92 | 102 | |
| Gemini 3.0 Flash2026.04 | 45.8 | 95.8 | 33.3 | 83.3 | 0.53 | 114 | |
| Gemini 3.1 Pro2026.04 | 37.5 | 58.3 | 47.1 | 76.5 | 2.31 | 43 | |
| Sonnet 4.62026.04 | 33.3 | 58.3 | 23.3 | 60 | 3.64 | 58 |