Sabotage Detection on ImpossibleBench
100Average PrecisionMonitor
Evaluation Results
| Method | Links | |
|---|---|---|
| MonitorTPC=100, Model=Qwen3.52026.04 | 100 | |
| BayesianTPC=100, Model=Qwen3.52026.04 | 100 | |
| BayesianTPC=100, Model=GPT-5.4m2026.04 | 100 | |
| BufferTPC=100, Model=Qwen3.52026.04 | 99 | |
| BufferTPC=100, Model=GPT-5.4m2026.04 | 99 | |
| BayesianTPC=50, Model=Qwen3.52026.04 | 96.9 | |
| BayesianTPC=50, Model=GPT-5.4m2026.04 | 96.9 | |
| BufferTPC=50, Model=Qwen3.52026.04 | 96.5 | |
| BufferTPC=50, Model=GPT-5.4m2026.04 | 96.5 | |
| MeerkatTPC=100, Model=Qwen3.52026.04 | 96.2 | |
| Naive AgentTPC=100, Model=GPT-5.4m2026.04 | 94.2 | |
| MonitorTPC=50, Model=Qwen3.52026.04 | 94 | |
| MeerkatTPC=50, Model=Qwen3.52026.04 | 92.7 | |
| BayesianTPC=10, Model=Qwen3.52026.04 | 92.6 | |
| BayesianTPC=10, Model=GPT-5.4m2026.04 | 92.5 | |
| MeerkatTPC=10, Model=Qwen3.52026.04 | 92 | |
| BufferTPC=10, Model=Qwen3.52026.04 | 90 | |
| BufferTPC=10, Model=GPT-5.4m2026.04 | 90 | |
| BayesianTPC=25, Model=GPT-5.4m2026.04 | 88.2 | |
| BayesianTPC=25, Model=Qwen3.52026.04 | 88.1 | |
| MeerkatTPC=10, Model=GPT-5.4m2026.04 | 86.9 | |
| BufferTPC=25, Model=Qwen3.52026.04 | 86.7 | |
| BufferTPC=25, Model=GPT-5.4m2026.04 | 86.7 | |
| MeerkatTPC=25, Model=Qwen3.52026.04 | 85.2 | |
| MonitorTPC=100, Model=GPT-5.4m2026.04 | 84.8 | |
| MonitorTPC=10, Model=GPT-5.4m2026.04 | 83.4 | |
| MeerkatTPC=100, Model=GPT-5.4m2026.04 | 83.3 | |
| Naive AgentTPC=50, Model=Qwen3.52026.04 | 82.4 | |
| MonitorTPC=10, Model=Qwen3.52026.04 | 80 | |
| Naive AgentTPC=100, Model=Qwen3.52026.04 | 78.9 | |
| MonitorTPC=50, Model=GPT-5.4m2026.04 | 78.6 | |
| MonitorTPC=25, Model=Qwen3.52026.04 | 78 | |
| Naive AgentTPC=10, Model=Qwen3.52026.04 | 77.7 | |
| MeerkatTPC=25, Model=GPT-5.4m2026.04 | 74.9 | |
| MonitorTPC=25, Model=GPT-5.4m2026.04 | 74.5 | |
| MeerkatTPC=50, Model=GPT-5.4m2026.04 | 73.6 | |
| Naive AgentTPC=50, Model=GPT-5.4m2026.04 | 68.7 | |
| Naive AgentTPC=25, Model=Qwen3.52026.04 | 62.6 | |
| Naive AgentTPC=10, Model=GPT-5.4m2026.04 | 50 | |
| Naive AgentTPC=25, Model=GPT-5.4m2026.04 | 50 |