Evaluation profile
MT-JailBench CrescendoX
1sub-evals
0.0213%total index weight
1components
Within-component eval weight: Misuse resistance 0.213%.
Model score (higher is better)Predicted score
About this eval
Multi-turn jailbreak resistance under fixed CrescendoX conditions.
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction | Total index weight | Within-component weight |
|---|---|---|---|---|
| safety_scoremt-jailbench/mt-jailbench.csv:safety_scoreMeasures resistance to the fixed CrescendoX multi-turn jailbreak protocol. | ordinary_harm_misuse_resistance:1.000harmbench | Higher is better | 0.0213% | Misuse resistance 0.213% |
safety_score
Measures resistance to the fixed CrescendoX multi-turn jailbreak protocol.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-haiku-4.5 | 81.76 | official | |
| 2 | claude-sonnet-4.5 | 75.47 | official | |
| 3 | gpt-5 | 61.29 | official | |
| 4 | qwen3.5-plus | 47.8 | official | |
| 5 | nemotron-3-super-120b-a12b | 44.03 | official | |
| 6 | gemini-3-pro-preview | 36.48 | official | |
| 7 | qwen3.5-flash | 34.59 | official | |
| 8 | llama-3-8b-instruct | 28.93 | official | |
| 9 | gemma-4-26b-a4b-it | 26.58 | official | |
| 10 | llama-4-maverick | 22.64 | official | |
| 11 | gemma-4-e4b | 21.38 | official | |
| 12 | grok-4.1-fast | 18.87 | official | |
| 13 | deepseek-v3.2 | 17.61 | official | |
| 14 | qwen3-4b-2507-instruct | 16.35 | official | |
| 15 | gpt-4o | 15.72 | official | |
| 16 | gemini-3-flash-preview | 13.84 | official | |
| 17 | llama-3-70b-instruct | 11.32 | official | |
| 17 | nemotron-3-nano-30b-a3b | 11.32 | official | |
| 17 | qwen3-30b-a3b-instruct | 11.32 | official | |
| 20 | llama-4-scout | 10.06 | official | |
| 21 | ministral-3-8b | 0.63 | official |