Rule-based ethics review on ethics-review benchmark rule-based
94.44RecallMirror
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MirrorSetting=Mirror-Agent, Model Parameters=8B2026.02 | 94.44 | 76.4 | 84.47 | 79.17 | |
| GPT-4.1Setting=Mirror-Agent2026.02 | 63.19 | 67.91 | 65.47 | 40 | |
| Qwen3-8BSetting=Mirror-Agent, Model Parameters=8B2026.02 | 55.23 | 60.63 | 57.8 | 25 | |
| DeepSeek-V3Setting=Mirror-Agent, Model Parameters=671B2026.02 | 45.83 | 84.62 | 59.46 | 20 | |
| DeepSeek-V3Setting=Zero-shot, Model Parameters=671B2026.02 | 18.06 | 33.33 | 23.42 | 20 | |
| GPT-4.1Setting=Zero-shot2026.02 | 15.28 | 22 | 18.03 | 40.83 | |
| Qwen3-8BSetting=Zero-shot, Model Parameters=8B2026.02 | 12.31 | 18.42 | 14.76 | 15 |