List your company
ResearchFormerly WizWand

Jailbreaking on HarmBench n = 200

16ASRGCG
13.1432.44551.7571.055Jul 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GCGTarget Model=Qwen3-8B, Iterations=@82026.0716
GCGTarget Model=GLM-4-9B, Iterations=@82026.0719
GCGTarget Model=Llama-3-8B, Iterations=@82026.0719.5
CSRTTarget Model=DeepSeek-R1, Iterations=@82026.0732.5
GCGTarget Model=DeepSeek-R1, Iterations=@82026.0741.5
CSRTTarget Model=Qwen3-8B, Iterations=@82026.0749.5
CSRTTarget Model=Llama-3-8B, Iterations=@82026.0760.5
CSRTTarget Model=Gemma-7B, Iterations=@82026.0766
STEERTarget Model=DeepSeek-R1, Iterations=@82026.0766
GCGTarget Model=Gemma-7B, Iterations=@82026.0773
STEERTarget Model=Qwen3-8B, Iterations=@82026.0776
STEERTarget Model=Llama-3-8B, Iterations=@82026.0776.5
CSRTTarget Model=Mistral-7B, Iterations=@82026.0778.5
CSRTTarget Model=GLM-4-9B, Iterations=@82026.0781.5
GCGTarget Model=Mistral-7B, Iterations=@82026.0782.5
STEERTarget Model=Gemma-7B, Iterations=@82026.0786
STEERTarget Model=GLM-4-9B, Iterations=@82026.0786
STEERTarget Model=Mistral-7B, Iterations=@82026.0787.5