Average Number of Queries for Jailbreak on AdvBench
2.1Avg QueriesLATS
Evaluation Results
| Method | Links | |
|---|---|---|
| LATSTarget Model=GPT-5.12026.01 | 2.1 | |
| LATSTarget Model=Vicuna-13B2026.01 | 2.1 | |
| Cres.Target Model=Mistral-7B2026.01 | 2.4 | |
| LATSTarget Model=GPT-4o2026.01 | 3.9 | |
| LATSTarget Model=Mistral-7B2026.01 | 3.9 | |
| LATSTarget Model=Mistral-7B-Inst2026.01 | 4.1 | |
| Cres.Target Model=Vicuna-13B2026.01 | 4.3 | |
| LATSTarget Model=Llama-2-7B-chat2026.01 | 5.1 | |
| LATSTarget Model=Llama-3.1-8B-Inst2026.01 | 5.8 | |
| Cres.Target Model=Llama-3.1-8B-Inst2026.01 | 6.4 | |
| Cres.Target Model=Mistral-7B-Inst2026.01 | 6.5 | |
| LATSTarget Model=Claude-Sonnet-4.52026.01 | 7.7 | |
| Cres.Target Model=Llama-2-7B-chat2026.01 | 8.2 | |
| ActorTarget Model=GPT-4o2026.01 | 10.1 | |
| ActorTarget Model=Vicuna-13B2026.01 | 10.1 | |
| ActorTarget Model=Mistral-7B2026.01 | 10.1 | |
| ActorTarget Model=Claude-Sonnet-4.52026.01 | 10.2 | |
| Cres.Target Model=Claude-Opus-4.52026.01 | 10.2 | |
| ActorTarget Model=Llama-3.1-8B-Inst2026.01 | 10.2 | |
| ActorTarget Model=Mistral-7B-Inst2026.01 | 10.2 | |
| ActorTarget Model=GPT-5.12026.01 | 10.3 | |
| ActorTarget Model=Llama-2-7B-chat2026.01 | 10.3 | |
| ActorTarget Model=Claude-Opus-4.52026.01 | 10.4 | |
| Cres.Target Model=Claude-Sonnet-4.52026.01 | 10.6 | |
| Cres.Target Model=GPT-4o2026.01 | 11.3 | |
| Cres.Target Model=GPT-5.12026.01 | 12.1 | |
| TAPTarget Model=Mistral-7B2026.01 | 12.3 | |
| SlotGCGModel=Mistral-7B, Base Attack=AttnGCG2026.06 | 12.34 | |
| SlotGCGModel=Qwen-2.5, Base Attack=I-GCG2026.06 | 12.63 | |
| SlotGCGModel=Mistral-7B, Base Attack=GCG-Hij2026.06 | 12.74 | |
| CoATarget Model=Mistral-7B2026.01 | 14.3 | |
| CoATarget Model=GPT-5.12026.01 | 14.5 | |
| CoATarget Model=Mistral-7B-Inst2026.01 | 14.6 | |
| CoATarget Model=Vicuna-13B2026.01 | 15.2 | |
| TAPTarget Model=Mistral-7B-Inst2026.01 | 15.6 | |
| SlotGCGModel=Llama-3.1-8B, Base Attack=GCG-Hij2026.06 | 15.65 | |
| CoATarget Model=Llama-3.1-8B-Inst2026.01 | 16.3 | |
| CoATarget Model=Claude-Sonnet-4.52026.01 | 16.4 | |
| SlotGCGModel=Llama-3.1-8B, Base Attack=AttnGCG2026.06 | 16.53 | |
| GCG-HijModel=Mistral-7B, Configuration=Base2026.06 | 17.2 | |
| SlotGCGModel=Mistral-7B, Base Attack=I-GCG2026.06 | 17.32 | |
| CoATarget Model=GPT-4o2026.01 | 18.2 | |
| TAPTarget Model=Llama-2-7B-chat2026.01 | 18.4 | |
| CoATarget Model=Llama-2-7B-chat2026.01 | 18.5 | |
| CoATarget Model=Claude-Opus-4.52026.01 | 18.7 | |
| TAPTarget Model=Vicuna-13B2026.01 | 18.7 | |
| I-GCGModel=Qwen-2.5, Configuration=Base2026.06 | 18.86 | |
| SlotGCGModel=Vicuna-7B, Base Attack=AttnGCG2026.06 | 18.96 | |
| SlotGCGModel=Llama-2-7B, Base Attack=I-GCG2026.06 | 19.14 | |
| SlotGCGModel=Llama-3.1-8B, Base Attack=GCG2026.06 | 19.29 | |
| SlotGCGModel=Mistral-7B, Base Attack=GCG2026.06 | 19.34 | |
| SlotGCGModel=Average, Base Attack=I-GCG2026.06 | 20.11 | |
| SlotGCGModel=Average, Base Attack=AttnGCG2026.06 | 20.21 | |
| PAIRTarget Model=Vicuna-13B2026.01 | 20.3 | |
| PAIRTarget Model=Mistral-7B2026.01 | 20.4 | |
| PAIRTarget Model=Llama-3.1-8B-Inst2026.01 | 20.5 | |
| PAIRTarget Model=Mistral-7B-Inst2026.01 | 20.8 | |
| I-GCGModel=Mistral-7B, Configuration=Base2026.06 | 21.08 | |
| SlotGCGModel=Llama-2-13B, Base Attack=AttnGCG2026.06 | 21.53 | |
| LATSTarget Model=Claude-Opus-4.52026.01 | 22.6 | |
| TAPTarget Model=Llama-3.1-8B-Inst2026.01 | 22.6 | |
| GCGModel=Vicuna-7B, Configuration=Base2026.06 | 22.85 | |
| SlotGCGModel=Llama-2-13B, Base Attack=I-GCG2026.06 | 23.02 | |
| SlotGCGModel=Vicuna-7B, Base Attack=I-GCG2026.06 | 23.16 | |
| SlotGCGModel=Vicuna-7B, Base Attack=GCG2026.06 | 23.61 | |
| GCGModel=Mistral-7B, Configuration=Base2026.06 | 25.16 | |
| SlotGCGModel=Average, Base Attack=GCG-Hij2026.06 | 25.25 | |
| SlotGCGModel=Llama-3.1-8B, Base Attack=I-GCG2026.06 | 25.39 | |
| SlotGCGModel=Vicuna-7B, Base Attack=GCG-Hij2026.06 | 25.52 | |
| SlotGCGModel=Qwen-2.5, Base Attack=AttnGCG2026.06 | 25.94 | |
| SlotGCGModel=Llama-2-7B, Base Attack=AttnGCG2026.06 | 25.98 | |
| TAPTarget Model=GPT-4o2026.01 | 26.8 | |
| AttnGCGModel=Vicuna-7B, Configuration=Base2026.06 | 27.49 | |
| SlotGCGModel=Qwen-2.5, Base Attack=GCG-Hij2026.06 | 27.84 | |
| PAIRTarget Model=Claude-Opus-4.52026.01 | 28.4 | |
| GCG-HijModel=Vicuna-7B, Configuration=Base2026.06 | 28.55 | |
| SlotGCGModel=Average, Base Attack=GCG2026.06 | 28.59 | |
| ADANTarget Model=Mistral-7B2026.01 | 28.6 | |
| I-GCGModel=Vicuna-7B, Configuration=Base2026.06 | 28.63 | |
| PAIRTarget Model=Claude-Sonnet-4.52026.01 | 28.7 | |
| GCGModel=Qwen-2.5, Configuration=Base2026.06 | 28.86 | |
| TAPTarget Model=GPT-5.12026.01 | 30.2 | |
| SlotGCGModel=Qwen-2.5, Base Attack=GCG2026.06 | 30.76 | |
| ADANTarget Model=Mistral-7B-Inst2026.01 | 32.4 | |
| AttnGCGModel=Mistral-7B, Configuration=Base2026.06 | 34.08 | |
| SlotGCGModel=Llama-2-13B, Base Attack=GCG-Hij2026.06 | 34.72 | |
| SlotGCGModel=Llama-2-7B, Base Attack=GCG-Hij2026.06 | 35.02 | |
| SlotGCGModel=Llama-2-13B, Base Attack=GCG2026.06 | 38.01 | |
| ADANTarget Model=Vicuna-13B2026.01 | 38.5 | |
| TAPTarget Model=Claude-Opus-4.52026.01 | 40.3 | |
| SlotGCGModel=Llama-2-7B, Base Attack=GCG2026.06 | 40.5 | |
| ADANTarget Model=Llama-2-7B-chat2026.01 | 40.6 | |
| TAPTarget Model=Claude-Sonnet-4.52026.01 | 44.5 | |
| ADANTarget Model=Llama-3.1-8B-Inst2026.01 | 46.3 | |
| GCG-HijModel=Llama-3.1-8B, Configuration=Base2026.06 | 48.1 | |
| PAIRTarget Model=GPT-5.12026.01 | 48.3 | |
| PAIRTarget Model=GPT-4o2026.01 | 48.6 | |
| ADANTarget Model=GPT-4o2026.01 | 56.4 | |
| GCG-HijModel=Average, Configuration=Base2026.06 | 59.65 | |
| ADANTarget Model=GPT-5.12026.01 | 60.7 |