Safety Evaluation on MMSafetyBench (Refusal Rate)
99.9Refusal RateSASA
Evaluation Results
| Method | Links | |
|---|---|---|
| SASAModel=GEMMA32026.06 | 99.9 | |
| SASAModel=QWEN3-VL2026.06 | 99.6 | |
| SASAModel=QWEN3.52026.06 | 99.3 | |
| SASAModel=INTERNVL 3.52026.06 | 99.1 | |
| SASAModel=MOLMO22026.06 | 98.4 | |
| MARSModel=QWEN3-VL2026.06 | 85.3 | |
| MARSModel=MOLMO22026.06 | 81.3 | |
| AdaSteerModel=QWEN3.52026.06 | 80 | |
| MARSModel=QWEN3.52026.06 | 76.9 | |
| AdaSteerModel=QWEN3-VL2026.06 | 59.9 | |
| MARSModel=INTERNVL 3.52026.06 | 55.9 | |
| ECSOModel=MOLMO22026.06 | 46.5 | |
| Zero-shotModel=QWEN3-VL2026.06 | 46.3 | |
| ECSOModel=QWEN3.52026.06 | 45.5 | |
| Zero-shotModel=QWEN3.52026.06 | 45.2 | |
| AdaSteerModel=MOLMO22026.06 | 45.2 | |
| ECSOModel=QWEN3-VL2026.06 | 44.8 | |
| Zero-shotModel=MOLMO22026.06 | 39.8 | |
| MARSModel=GEMMA32026.06 | 32.5 | |
| AdaSteerModel=INTERNVL 3.52026.06 | 32.3 | |
| Zero-shotModel=INTERNVL 3.52026.06 | 26.3 | |
| ECSOModel=INTERNVL 3.52026.06 | 26.3 | |
| Zero-shotModel=GEMMA32026.06 | 23.7 | |
| ECSOModel=GEMMA32026.06 | 23.7 | |
| AdaSteerModel=GEMMA32026.06 | 21.3 |