Safety Evaluation on JailbreakV-28K v1 (test)
6.63ASR (Noise-T)RAI
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RAIModel=Qwen3-VL-8B2026.02 | 6.63 | 0 | 0 | 8.86 | 0 | 0 | 7.34 | 0 | 0 | 8.19 | 0 | 0 | 2.58 | |
| RAIModel=DeepSeek-VL-7B2026.02 | 6.83 | 0 | 1.35 | 5.1 | 0.29 | 6.33 | 6.15 | 0 | 1.35 | 6.5 | 0 | 1.35 | 2.94 | |
| shiftDCModel=LLaVA-1.6-7B2026.02 | 7.72 | 3.82 | 9.13 | 12.13 | 2.65 | 7.86 | 13.65 | 2.48 | 8.5 | 12.41 | 2.43 | 7.6 | 8.36 | |
| shiftDCModel=Qwen3-VL-8B2026.02 | 9.22 | 0.46 | 0.85 | 13.46 | 1.22 | 0 | 8.33 | 2.53 | 1 | 20.56 | 1.25 | 2.06 | 5.07 | |
| shiftDCModel=DeepSeek-VL-7B2026.02 | 10.52 | 7.66 | 5.16 | 15.48 | 5 | 3.93 | 14.25 | 1.1 | 1.85 | 16.12 | 3.45 | 4.93 | 7.45 | |
| RAIModel=LLaVA-1.6-7B2026.02 | 10.66 | 0 | 0 | 8.8 | 0 | 0 | 7.65 | 0 | 0 | 6.63 | 0 | 0 | 2.81 | |
| AdashieldModel=Qwen3-VL-8B2026.02 | 12.42 | 0.65 | 0.73 | 12.33 | 1.3 | 0 | 9.25 | 1.55 | 0.63 | 14.36 | 0.71 | 1.22 | 4.59 | |
| ECSOModel=DeepSeek-VL-7B2026.02 | 20.12 | 9.55 | 10.46 | 30.05 | 6.5 | 5 | 34.33 | 1.5 | 2.66 | 38.45 | 6.74 | 10.44 | 14.65 | |
| AdashieldModel=LLaVA-1.6-7B2026.02 | 20.62 | 5.14 | 14.29 | 19.33 | 4.12 | 12.11 | 21.1 | 4.6 | 13.5 | 20.74 | 3.59 | 12.06 | 12.6 | |
| AdashieldModel=DeepSeek-VL-7B2026.02 | 21.6 | 8.93 | 8.93 | 14.2 | 6.72 | 2.7 | 22.73 | 2.59 | 3.11 | 19.6 | 7.34 | 5.33 | 10.32 | |
| CoCAModel=DeepSeek-VL-7B2026.02 | 22.16 | 9.66 | 12.53 | 36.14 | 6.48 | 7.92 | 35.13 | 2.83 | 3.6 | 40.2 | 8.1 | 12 | 16.4 | |
| OriginalModel=DeepSeek-VL-7B2026.02 | 24.23 | 9.65 | 14.86 | 48.65 | 7.31 | 9.46 | 37.13 | 3.15 | 4.05 | 47.93 | 9.65 | 14.86 | 19.24 | |
| COCAModel=Qwen3-VL-8B2026.02 | 52.1 | 2.55 | 2.4 | 48.55 | 1.35 | 0 | 54.3 | 2.66 | 1.2 | 46.23 | 2.11 | 3.52 | 18.08 | |
| ECSOModel=Qwen3-VL-8B2026.02 | 53.01 | 2.25 | 2.36 | 49.85 | 1 | 0 | 55.12 | 2.77 | 1.25 | 47 | 1.35 | 1.66 | 18.16 | |
| OriginalModel=Qwen3-VL-8B2026.02 | 54.74 | 2.33 | 2.7 | 51.13 | 1.46 | 0 | 56.87 | 2.63 | 1.35 | 48.32 | 1.46 | 2.77 | 18.81 | |
| COCAModel=LLaVA-1.6-7B2026.02 | 79.25 | 19.56 | 54.33 | 77.47 | 14.33 | 45.26 | 78.19 | 14.82 | 46.44 | 79.55 | 11.01 | 43.1 | 46.94 | |
| ECSOModel=LLaVA-1.6-7B2026.02 | 80.88 | 16.59 | 55.22 | 78.25 | 14.66 | 46.15 | 79.82 | 15.16 | 47.14 | 77.62 | 12.46 | 42.8 | 47.22 | |
| OriginalModel=LLaVA-1.6-7B2026.02 | 82.5 | 20.76 | 56.76 | 80.8 | 15.2 | 47.3 | 81.24 | 15.79 | 48.65 | 82 | 12.87 | 45.95 | 49.15 |