Safety Evaluation on HarmfulQA and JailbreakBench
0.046Mean Danger ScoreBaseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineModel=Claude-4.5-Sonnet2026.02 | 0.046 | |
| BaselineModel=Qwen-3-8b2026.02 | 0.09 | |
| MIPROv2Model=Claude-4.5-Sonnet2026.02 | 0.103 | |
| BaselineModel=Llama-4-maverick2026.02 | 0.215 | |
| GEPAModel=Claude-4.5-Sonnet2026.02 | 0.236 | |
| SIMBAModel=Claude-4.5-Sonnet2026.02 | 0.347 | |
| GEPAModel=Llama-4-maverick2026.02 | 0.469 | |
| GEPAModel=Qwen-3-8b2026.02 | 0.477 | |
| MIPROv2Model=Llama-4-maverick2026.02 | 0.581 | |
| SIMBAModel=Llama-4-maverick2026.02 | 0.623 | |
| BaselineModel=Gemini-2.5-pro2026.02 | 0.645 | |
| MIPROv2Model=Gemini-2.5-pro2026.02 | 0.704 | |
| GEPAModel=Gemini-2.5-pro2026.02 | 0.734 | |
| MIPROv2Model=Qwen-3-8b2026.02 | 0.746 | |
| SIMBAModel=Gemini-2.5-pro2026.02 | 0.774 | |
| SIMBAModel=Qwen-3-8b2026.02 | 0.792 |