Safety Classification on CoSApien (test)
94.6F1 ScoreCSRM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CSRMseverity augmentation=true, configurable safety configuration augmentation=true2026.05 | 94.6 | 98.7 | 0.076 | |
| ShieldGemma-9B2026.05 | 86.9 | 92.5 | 0.156 | |
| OSS-Safeguard-20B-High2026.05 | 85.9 | 88.4 | 0.177 | |
| LlamaGuard-3-8B2026.05 | 84 | 95.3 | 0.169 | |
| Llama-3.1-70B-InstChain-of-Thought=true2026.05 | 82.7 | 87.5 | 0.249 | |
| CSRMseverity augmentation=false, configurable safety configuration augmentation=false2026.05 | 82.4 | 88.3 | 0.141 | |
| Qwen3-30BThinking Mode=true2026.05 | 82.2 | 90.9 | 0.21 | |
| CSRMseverity augmentation=true, configurable safety configuration augmentation=false2026.05 | 81.5 | 88.6 | 0.123 | |
| Llama-3.1-8B-Inst2026.05 | 81.4 | 97.3 | 0.169 | |
| DynaGuard-8BChain-of-Thought=true2026.05 | 78.2 | 83.3 | 0.196 |