Refusal Suppression on AdvBench
3.33Targeted Content Rate (TCR)Original
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OriginalBase model=Kimi-VL-A3B2026.06 | 3.33 | 60.15 | |
| OriginalBase model=DeepSeek-VL2-small2026.06 | 5 | 57.03 | |
| OriginalBase model=Qwen3-VL-30B-A3B2026.06 | 11.32 | 71.79 | |
| SOMBase model=Qwen3-VL-30B-A3B2026.06 | 16.98 | 68.96 | |
| OriginalBase model=InternVL3.5-30B-A3B2026.06 | 18.24 | 66.12 | |
| GateBreakerBase model=Kimi-VL-A3B2026.06 | 30.82 | 49.35 | |
| SOMBase model=Kimi-VL-A3B2026.06 | 48.43 | 60.16 | |
| HereticBase model=InternVL3.5-30B-A3B2026.06 | 55.83 | 66.29 | |
| GateBreakerBase model=InternVL3.5-30B-A3B2026.06 | 57.23 | 63.1 | |
| HereticBase model=Qwen3-VL-30B-A3B2026.06 | 60.83 | 71.75 | |
| GateBreakerBase model=Qwen3-VL-30B-A3B2026.06 | 67.92 | 73.21 | |
| SOMBase model=DeepSeek-VL2-small2026.06 | 69.18 | 57.97 | |
| GateBreakerBase model=DeepSeek-VL2-small2026.06 | 69.18 | 57.21 | |
| HereticBase model=DeepSeek-VL2-small2026.06 | 75 | 56.45 | |
| SOMBase model=InternVL3.5-30B-A3B2026.06 | 82.39 | 68.11 | |
| HereticBase model=Kimi-VL-A3B2026.06 | 85 | 59.03 | |
| LoMCBase model=Qwen3-VL-30B-A3B2026.06 | 89.17 | 72.23 | |
| LoMCBase model=DeepSeek-VL2-small2026.06 | 90 | 57.75 | |
| LoMCBase model=Kimi-VL-A3B2026.06 | 93.33 | 60.28 | |
| LoMCBase model=InternVL3.5-30B-A3B2026.06 | 97.5 | 66.46 |