Refusal Suppression on VLSBench
100TCRLoMC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LoMCBase model=Qwen3-VL-30B-A3B2026.06 | 100 | 72.12 | |
| HereticBase model=Qwen3-VL-30B-A3B2026.06 | 99.87 | 71.67 | |
| LoMCBase model=InternVL3.5-30B-A3B2026.06 | 99.2 | 67.01 | |
| LoMCBase model=DeepSeek-VL2-small2026.06 | 98.93 | 57.64 | |
| HereticBase model=DeepSeek-VL2-small2026.06 | 98.27 | 57.29 | |
| HereticBase model=InternVL3.5-30B-A3B2026.06 | 97.6 | 66.17 | |
| LoMCBase model=Kimi-VL-A3B2026.06 | 97.6 | 60.34 | |
| SOMBase model=InternVL3.5-30B-A3B2026.06 | 76.53 | 67.84 | |
| SOMBase model=DeepSeek-VL2-small2026.06 | 69.2 | 58.72 | |
| SOMBase model=Kimi-VL-A3B2026.06 | 67.07 | 60.86 | |
| GateBreakerBase model=DeepSeek-VL2-small2026.06 | 65.41 | 57.46 | |
| GateBreakerBase model=Qwen3-VL-30B-A3B2026.06 | 64.41 | 73.8 | |
| SOMBase model=Qwen3-VL-30B-A3B2026.06 | 56.8 | 65.27 | |
| GateBreakerBase model=InternVL3.5-30B-A3B2026.06 | 42.14 | 62.28 | |
| GateBreakerBase model=Kimi-VL-A3B2026.06 | 42.14 | 54.46 | |
| HereticBase model=Kimi-VL-A3B2026.06 | 15.6 | 59.67 | |
| OriginalBase model=DeepSeek-VL2-small2026.06 | 1.87 | 57.03 | |
| OriginalBase model=Qwen3-VL-30B-A3B2026.06 | 1.33 | 71.79 | |
| OriginalBase model=InternVL3.5-30B-A3B2026.06 | 1.33 | 66.12 | |
| OriginalBase model=Kimi-VL-A3B2026.06 | 1.07 | 60.15 |