List your company

Jailbreak Attack on Advbench-M

0Attack Success Rate (ASR%)Vanilla
-3.68421.18346.0570.917Mar 8, 2026Mar 14, 2026Mar 20, 2026Mar 26, 2026Apr 1, 2026Apr 7, 2026Apr 14, 2026

Evaluation results

MethodLinks
VanillaTarget Model=GPT-4o2026.0300
FigStepTarget Model=GPT-4o2026.0300.8
Qwen3-VL-Plus2026.040—
Gemini-3-Flash2026.040—
Claude-Haiku-4.52026.040—
GPT-5-Mini2026.040—
Qwen3-VL-8B-InstructParameters=8B, Alignment=Instruct2026.040—
DeepSeek-V3.22026.040.14—
Kimi-K2.52026.040.41—
VanillaTarget Model=Gemini-2.0-Flash2026.030.51.3
VanillaTarget Model=Gemini-2.5-Flash2026.030.51.3
FigStepTarget Model=Gemini-2.5-Flash2026.030.91.8
VanillaTarget Model=Qwen3-VL-Flash2026.030.91.5
Mistral-Medium-32026.041.1—
Llama-3.2-11B-VisionParameters=11B, Type=Vision2026.041.51—
VanillaTarget Model=Qwen2.5VL-7B2026.031.91.6
SI-AttackTarget Model=InternVL-3-9B2026.031.92.4
VanillaTarget Model=InternVL-3-9B2026.032.31.3
FigStepTarget Model=Qwen2.5VL-7B2026.032.83
FigStepTarget Model=Qwen3-VL-Flash2026.033.22.9
SI-AttackTarget Model=Qwen3-VL-Flash2026.033.72.8
HADESTarget Model=Gemini-2.0-Flash2026.035.11.7
JOODTarget Model=Gemini-2.5-Flash2026.035.11.9
HADESTarget Model=Qwen3-VL-Flash2026.035.12.1
FigStepTarget Model=Gemini-2.0-Flash2026.037.42.8
HADESTarget Model=Gemini-2.5-Flash2026.037.91.8
FigStep-ProTarget Model=Gemini-2.5-Flash2026.038.32.5
HADESTarget Model=GPT-4o2026.0310.21.5
FigStep-ProTarget Model=GPT-4o2026.0310.71.7
SI-AttackTarget Model=Qwen2.5VL-7B2026.0311.12.9
FigStepTarget Model=InternVL-3-9B2026.0315.72.4
JOODTarget Model=Gemini-2.0-Flash2026.0316.22.3
LLaVA-v1.6-Vicuna-7BBackbone=Vicuna-7B2026.0417.56—
JOODTarget Model=Qwen3-VL-Flash2026.0318.52.9
JOODTarget Model=GPT-4o2026.0319.92
SI-AttackTarget Model=Gemini-2.5-Flash2026.0321.32.3
HADESTarget Model=Qwen2.5VL-7B2026.0322.73.3
HADESTarget Model=InternVL-3-9B2026.0324.52.1
SI-AttackTarget Model=GPT-4o2026.0332.92.5
JOODTarget Model=Qwen2.5VL-7B2026.0333.33.1
SI-AttackTarget Model=Gemini-2.0-Flash2026.0347.72.8
FigStep-ProTarget Model=Qwen3-VL-Flash2026.0348.64.1
StructAttack (v2)Target Model=Gemini-2.5-Flash2026.0349.15.4
StructAttack (v3)Target Model=Gemini-2.5-Flash2026.0349.55.8
StructAttack (v1)Target Model=Gemini-2.5-Flash2026.0352.36.1
JOODTarget Model=InternVL-3-9B2026.0354.23.4
StructAttack (v1)Target Model=Qwen3-VL-Flash2026.0359.35.9
FigStep-ProTarget Model=Gemini-2.0-Flash2026.0362.55.8
FigStep-ProTarget Model=Qwen2.5VL-7B2026.0364.45
StructAttack (v3)Target Model=GPT-4o2026.0365.35.5
StructAttack (v1)Target Model=GPT-4o2026.03695.3
StructAttack (v2)Target Model=GPT-4o2026.03695.6
StructAttack (v1)Target Model=Gemini-2.0-Flash2026.0369.97
StructAttack (v3)Target Model=Gemini-2.0-Flash2026.0370.46.7
StructAttack (v3)Target Model=Qwen3-VL-Flash2026.0370.46.8
StructAttack (v2)Target Model=Qwen3-VL-Flash2026.0370.86.6
StructAttack (v2)Target Model=Gemini-2.0-Flash2026.0373.66.5
StructAttack (v2)Target Model=Qwen2.5VL-7B2026.0384.76
FigStep-ProTarget Model=InternVL-3-9B2026.0386.15.5
StructAttack (v3)Target Model=Qwen2.5VL-7B2026.0386.66.4
StructAttack (v1)Target Model=InternVL-3-9B2026.03875.8
StructAttack (v1)Target Model=Qwen2.5VL-7B2026.0388.46.7
StructAttack (v3)Target Model=InternVL-3-9B2026.0389.85.1
StructAttack (v2)Target Model=InternVL-3-9B2026.0392.15.4