Share your thoughts, 1 month free Claude Pro on usSee more

Jailbreak Attack Evaluation on SafetyBench MCV

79.79ASR (1-Clip)

Qwen2.5-VL-32B

Updated 1d ago

Evaluation Results

Method	Links
Qwen2.5-VL-32B 2026.06		79.79	80.55	81.3	82.33	80.99	1.08
Qwen2.5-VL-72B 2026.06		74.52	76.03	75.07	76.1	75.43	0.77
Qwen3-VL-8B 2026.06		72.4	73.01	71.71	73.15	72.57	0.66
Qwen2.5-VL-32B 2026.06		71.71	79.18	80.68	81.1	78.17	4.38
Qwen2.5-VL-7B 2026.06		69.04	80	79.59	80.27	77.23	5.46
LLaVA-Video-7B 2026.06		66.58	66.23	65.62	66.85	66.32	0.53
InternVL3.5-38B 2026.06		64.11	65.34	66.44	67.05	65.74	1.29
InternVL3.5-8B 2026.06		64.04	64.86	64.93	65.27	64.78	0.52
Qwen3-VL-32B 2026.06		63.84	64.38	65.68	67.6	65.38	1.67
Qwen3-VL-32B 2026.06		59.86	61.99	62.88	63.97	62.18	1.74
InternVL3.5-38B 2026.06		59.04	59.73	61.3	62.74	60.7	1.66
Qwen3-VL-8B 2026.06		55.48	55.07	56.64	57.4	56.15	1.07
Qwen2.5-VL-7B 2026.06		50.75	61.23	66.78	68.7	61.87	8.06
LLaVA-Video-7B 2026.06		49.86	50	48.9	50.68	49.86	0.73
InternVL3.5-8B 2026.06		46.16	46.51	55.27	58.08	51.51	6.08
Qwen2.5-VL-72B 2026.06		43.7	47.53	55.82	57.6	51.16	6.63