Safety evaluation on ViSU unsafe 24
90Refusal RateSASA
Evaluation Results
| Method | Links | |
|---|---|---|
| SASAModel=GEMMA32026.06 | 90 | |
| MARSModel=QWEN3-VL2026.06 | 78.5 | |
| MARSModel=INTERNVL 3.52026.06 | 70.4 | |
| MARSModel=MOLMO22026.06 | 68.1 | |
| SASAModel=QWEN3.52026.06 | 65.1 | |
| SASAModel=MOLMO22026.06 | 64.2 | |
| SASAModel=INTERNVL 3.52026.06 | 60.2 | |
| MARSModel=QWEN3.52026.06 | 51.1 | |
| MARSModel=GEMMA32026.06 | 43.3 | |
| ECSOModel=MOLMO22026.06 | 37 | |
| AdaSteerModel=QWEN3.52026.06 | 23.9 | |
| Zero-shotModel=QWEN3.52026.06 | 22.9 | |
| ECSOModel=QWEN3.52026.06 | 22.9 | |
| SASAModel=QWEN3-VL2026.06 | 20.1 | |
| AdaSteerModel=GEMMA32026.06 | 18.7 | |
| AdaSteerModel=QWEN3-VL2026.06 | 18.6 | |
| Zero-shotModel=GEMMA32026.06 | 18.2 | |
| ECSOModel=GEMMA32026.06 | 17.7 | |
| ECSOModel=INTERNVL 3.52026.06 | 16.1 | |
| ECSOModel=QWEN3-VL2026.06 | 14.8 | |
| Zero-shotModel=QWEN3-VL2026.06 | 14.2 | |
| AdaSteerModel=INTERNVL 3.52026.06 | 13.6 | |
| Zero-shotModel=INTERNVL 3.52026.06 | 11.2 | |
| Zero-shotModel=MOLMO22026.06 | 0.2 | |
| AdaSteerModel=MOLMO22026.06 | 0.2 |