Security Interception on Bilingual Security Benchmark English
98.5IRQwen2.5-14B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2.5-14BCondition=Qwen2.5-14B2026.03 | 98.5 | 20.1 | 88.2 | 100 | 98.3 | 97.1 | 88.3 | |
| Llama-3.1-8BCondition=Llama-3.1-8B2026.03 | 97.8 | 37.7 | 76.1 | 98.6 | 97.2 | 97.8 | 80.8 | |
| GPT-4o-miniCondition=GPT-4o-mini2026.03 | 95.4 | 12.3 | 91.4 | 95 | 95.6 | 95.6 | 94.2 | |
| Qwen3.5-9BCondition=Qwen3.5-9B2026.03 | 95 | 23.7 | 84.6 | 96.5 | 92.8 | 96.3 | 77.5 | |
| Cascade (Q→14B)Condition=Cascade (Q→14B)2026.03 | 94.7 | 9.7 | 92.5 | 96.5 | 92.8 | 95.6 | — | |
| Qwen3.5-4BCondition=Qwen3.5-4B2026.03 | 94.3 | 29.2 | 80.9 | 94.3 | 96.1 | 91.9 | 83.3 | |
| Cascade (Q→G)Condition=Cascade (Q→G)2026.03 | 92.6 | 6.7 | 92.9 | 91.5 | 91.1 | 95.6 | — | |
| BART-MNLICondition=BART-MNLI2026.03 | 8.1 | 0 | 15 | 4.3 | 8.3 | 11.8 | 63.3 | |
| BaselineCondition=Baseline2026.03 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |