Model Vulnerability Evaluation on TRAP 1.0 (Overall) benchmark leaderboard · SOTA2 Research

Model Vulnerability Evaluation on TRAP 1.0 (Overall)

76.6Task Performance ScoreQwen3-VL-8B

Evaluation Results

Method	Links
Qwen3-VL-8B2026.06		76.6	9.4	16.7
Qwen3-VL-32B2026.06		76.2	8.8	15.8
GPT-5-mini2026.06		75.8	10	17.7
Devstral-Small-25122026.06		73	15.2	25.2
GPT-5.4-mini2026.06		72.6	10.4	18.2
GPT-52026.06		72.2	12.8	21.7
Gemini-2.5-Flash2026.06		71	10	17.5
InternVL3.5-14B2026.06		70.2	15.6	7.2
InternVL3.5-38B2026.06		69	12.8	6.2
InternVL3.5-8B2026.06		67.4	12.4	5
GPT-4o-mini2026.06		66.6	13.2	22
Gemini-2.5-Flash-Lite2026.06		66	20.4	31.2
Claude Sonnet 4.52026.06		62	6.6	11.9
Gemini-2.5-Pro2026.06		61.2	26.4	36.9
Claude Haiku 4.52026.06		60.4	8.8	15.4
Qwen3-VL-4B2026.06		58.6	30.8	40.4
InternVL3.5-2B2026.06		52.8	18.2	6.4
Phi-4-multimodal2026.06		33.6	38.4	35.8
Llama3.2-11B-Vision2026.06		31.2	22.2	25.9
InternVL3.5-4B2026.06		25.4	15	2.6
Qwen3-VL-2B2026.06		23.8	20.4	22
GLM-4.6V-Flash2026.06		22.6	17.4	19.7