Distributed Misuse Detection on DM-Bio 20x
59.3Trace-level APMeerkat
Evaluation Results
| Method | Links | |
|---|---|---|
| MeerkatModel=Qwen2.5-72B2026.04 | 59.3 | |
| Naive AgentModel=Qwen2.5-72B2026.04 | 57.2 | |
| BayesianModel=GPT-4o2026.04 | 24.6 | |
| BayesianModel=Qwen2.5-72B2026.04 | 17.6 | |
| BufferModel=Qwen2.5-72B2026.04 | 14.2 | |
| MeerkatModel=GPT-4o2026.04 | 10 | |
| BufferModel=GPT-4o2026.04 | 9.5 | |
| MonitorModel=Qwen2.5-72B2026.04 | 5.4 | |
| Naive AgentModel=GPT-4o2026.04 | 5 |