Refusal Suppression on StrongREJECT
100TCRLoMC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LoMCBase model=InternVL3.5-30B-A3B2026.06 | 100 | 67.07 | |
| LoMCBase model=Kimi-VL-A3B2026.06 | 98 | 60.46 | |
| LoMCBase model=Qwen3-VL-30B-A3B2026.06 | 94 | 72.3 | |
| LoMCBase model=DeepSeek-VL2-small2026.06 | 90 | 57.74 | |
| HereticBase model=Kimi-VL-A3B2026.06 | 79.33 | 58.63 | |
| SOMBase model=InternVL3.5-30B-A3B2026.06 | 77.96 | 68.33 | |
| SOMBase model=DeepSeek-VL2-small2026.06 | 77.96 | 58.67 | |
| HereticBase model=Qwen3-VL-30B-A3B2026.06 | 72.67 | 71.89 | |
| HereticBase model=DeepSeek-VL2-small2026.06 | 72 | 57.05 | |
| GateBreakerBase model=DeepSeek-VL2-small2026.06 | 69.33 | 56.11 | |
| HereticBase model=InternVL3.5-30B-A3B2026.06 | 57.33 | 66.33 | |
| GateBreakerBase model=InternVL3.5-30B-A3B2026.06 | 46.65 | 62.82 | |
| SOMBase model=Kimi-VL-A3B2026.06 | 45.05 | 62.37 | |
| GateBreakerBase model=Qwen3-VL-30B-A3B2026.06 | 29.39 | 69.51 | |
| GateBreakerBase model=Kimi-VL-A3B2026.06 | 28.43 | 57.4 | |
| SOMBase model=Qwen3-VL-30B-A3B2026.06 | 21.1 | 68.84 | |
| OriginalBase model=Kimi-VL-A3B2026.06 | 19.33 | 60.15 | |
| OriginalBase model=DeepSeek-VL2-small2026.06 | 14 | 57.03 | |
| OriginalBase model=InternVL3.5-30B-A3B2026.06 | 9.27 | 66.12 | |
| OriginalBase model=Qwen3-VL-30B-A3B2026.06 | 1.28 | 71.79 |