List your company

Jailbreak Attack on SafeBench (HF and ASR)

1.7HFFigStep-Pro
1.4882.9194.355.781Mar 8, 2026

Evaluation results

MethodLinks
FigStep-ProTarget Model=GPT-4o2026.031.74.9
SI-AttackTarget Model=GPT-4o2026.03217.7
VanillaTarget Model=Qwen3-VL-Flash2026.032.50.9
VanillaTarget Model=Qwen2.5VL-7B2026.032.51.1
VanillaTarget Model=Gemini-2.0-Flash2026.032.61.1
VanillaTarget Model=GPT-4o2026.032.82
FigStepTarget Model=GPT-4o2026.032.96.3
VanillaTarget Model=Gemini-2.5-Flash2026.0331.7
JOODTarget Model=GPT-4o2026.033.211.4
SI-AttackTarget Model=Gemini-2.0-Flash2026.033.751.4
HADESTarget Model=Qwen3-VL-Flash2026.033.85.4
SI-AttackTarget Model=Qwen2.5VL-7B2026.033.87.1
JOODTarget Model=Gemini-2.5-Flash2026.03419.7
VanillaTarget Model=InternVL-3-9B2026.03416.3
JOODTarget Model=Qwen3-VL-Flash2026.034.113.7
SI-AttackTarget Model=Qwen3-VL-Flash2026.034.324.6
SI-AttackTarget Model=InternVL-3-9B2026.034.335.7
HADESTarget Model=GPT-4o2026.034.420.6
FigStepTarget Model=Gemini-2.5-Flash2026.034.520.9
HADESTarget Model=Gemini-2.0-Flash2026.034.634
JOODTarget Model=Gemini-2.0-Flash2026.034.636.3
SI-AttackTarget Model=Gemini-2.5-Flash2026.034.628.6
FigStepTarget Model=Qwen3-VL-Flash2026.034.726.9
FigStepTarget Model=Qwen2.5VL-7B2026.034.746.6
HADESTarget Model=Gemini-2.5-Flash2026.034.825.4
FigStep-ProTarget Model=Gemini-2.5-Flash2026.034.934.4
FigStep-ProTarget Model=Qwen3-VL-Flash2026.035.238.9
FigStep-ProTarget Model=Qwen2.5VL-7B2026.035.452.9
HADESTarget Model=InternVL-3-9B2026.035.441.1
StructAttack (v3)Target Model=InternVL-3-9B2026.035.460.6
StructAttack (v3)Target Model=Qwen2.5VL-7B2026.035.674
JOODTarget Model=Qwen2.5VL-7B2026.035.840.6
HADESTarget Model=Qwen2.5VL-7B2026.035.936
FigStep-ProTarget Model=InternVL-3-9B2026.035.965.1
FigStepTarget Model=Gemini-2.0-Flash2026.03651.4
StructAttack (v2)Target Model=Gemini-2.5-Flash2026.03654
StructAttack (v2)Target Model=GPT-4o2026.036.155.7
StructAttack (v1)Target Model=Qwen2.5VL-7B2026.036.177.7
StructAttack (v2)Target Model=Qwen2.5VL-7B2026.036.180.6
JOODTarget Model=InternVL-3-9B2026.036.156.3
StructAttack (v2)Target Model=InternVL-3-9B2026.036.170.9
StructAttack (v1)Target Model=GPT-4o2026.036.356
StructAttack (v2)Target Model=Gemini-2.0-Flash2026.036.360
StructAttack (v1)Target Model=InternVL-3-9B2026.036.374.6
StructAttack (v3)Target Model=GPT-4o2026.036.456
StructAttack (v1)Target Model=Gemini-2.0-Flash2026.036.467.7
StructAttack (v1)Target Model=Gemini-2.5-Flash2026.036.450.9
StructAttack (v3)Target Model=Gemini-2.5-Flash2026.036.452.3
FigStepTarget Model=InternVL-3-9B2026.036.454
StructAttack (v3)Target Model=Gemini-2.0-Flash2026.036.559.7
StructAttack (v2)Target Model=Qwen3-VL-Flash2026.036.854
FigStep-ProTarget Model=Gemini-2.0-Flash2026.03763.1
StructAttack (v1)Target Model=Qwen3-VL-Flash2026.03756
StructAttack (v3)Target Model=Qwen3-VL-Flash2026.03756.9