Static Attack Detection on Static Benchmark 450-chain
84.88Position AccuracyReasoning Safety Monitor (Qwen3.5-35B-A3B)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reasoning Safety Monitor (Qwen3.5-35B-A3B)Monitor Backend=Qwen3.5-35B-A3B, configuration=taxonomy-embedded prompt2026.03 | 84.88 | 85.37 | |
| Reasoning Safety Monitor (Gemini-3-Flash-Preview)Monitor Backend=Gemini-3-Flash-Preview, configuration=taxonomy-embedded prompt2026.03 | 84 | 80.44 | |
| Reasoning Safety Monitor (gpt-oss-20b)Monitor Backend=gpt-oss-20b, configuration=taxonomy-embedded prompt2026.03 | 82.66 | 75.33 | |
| Reasoning Safety Monitor (GPT-4o-2024-08-06)Monitor Backend=GPT-4o-2024-08-06, configuration=taxonomy-embedded prompt2026.03 | 73.11 | 70 | |
| Qwen2.5-Math-PRM-7Bbaseline_type=process reward model (PRM)2026.03 | 68.83 | — | |
| SelfCheckGPTbaseline_type=hallucination detection2026.03 | 44.36 | — |