Jailbreak Robustness on Jailbreak Cipher, CodeChameleon (test)
95.75Cipher Success RateDASE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DASEBase Model=Kimi-K2-Instruct, Size=1T2026.01 | 95.75 | 60.9 | 78.33 | |
| Gemma-2 + SaRFTSize=9B2026.01 | 87.5 | 3.1 | 45.3 | |
| Kimi-K2-InstructSize=1T2026.01 | 85.5 | 39.5 | 62.5 | |
| gpt-oss-120bSize=120B2026.01 | 71 | 82.4 | 76.7 | |
| Qwen3-235B-A22B-Instruct-2507Size=235B2026.01 | 70.75 | 80.1 | 75.42 | |
| DASEBase Model=GPT-5.2, Size=Undisclosed2026.01 | 68 | 100 | 84 | |
| LLaMA-3 + SaRFTSize=8B2026.01 | 64 | 55.1 | 59.55 | |
| GPT-5.2Size=Undisclosed2026.01 | 63.5 | 95.6 | 79.55 | |
| Qwen3-32BSize=32B2026.01 | 61.75 | 55.8 | 58.77 | |
| Qwen2.5 + SaRFTSize=7B2026.01 | 30 | 14.2 | 22.1 |