Model Vulnerability Evaluation on TRAP Multimodal 1.0
73Task Performance ScoreGPT-5-mini
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5-miniModel Type=Proprietary2026.06 | 73 | 8 | |
| GPT-5Model Type=Proprietary2026.06 | 72 | 11.5 | |
| Qwen3-VL-8BModel Type=Open-source2026.06 | 71.5 | 8 | |
| Qwen3-VL-32BModel Type=Open-source2026.06 | 71 | 8 | |
| Devstral-Small-2512Model Type=Open-source2026.06 | 70.5 | 14.5 | |
| Gemini-2.5-FlashModel Type=Proprietary2026.06 | 69 | 9 | |
| InternVL3.5-38BModel Type=Open-source2026.06 | 68.5 | 12 | |
| GPT-5.4-miniModel Type=Proprietary2026.06 | 67.5 | 8 | |
| Gemini-2.5-Flash-LiteModel Type=Proprietary2026.06 | 66.5 | 18 | |
| InternVL3.5-14BModel Type=Open-source2026.06 | 66 | 12.5 | |
| InternVL3.5-8BModel Type=Open-source2026.06 | 64 | 12.5 | |
| GPT-4o-miniModel Type=Proprietary2026.06 | 60 | 13.5 | |
| Claude Haiku 4.5Model Type=Proprietary2026.06 | 55 | 6 | |
| Claude Sonnet 4.5Model Type=Proprietary2026.06 | 55 | 3.5 | |
| Gemini-2.5-ProModel Type=Proprietary2026.06 | 52 | 32 | |
| Qwen3-VL-4BModel Type=Open-source2026.06 | 51 | 27.5 | |
| InternVL3.5-2BModel Type=Open-source2026.06 | 44.5 | 17 | |
| Phi-4-multimodalModel Type=Open-source2026.06 | 16.5 | 48.5 | |
| Qwen3-VL-2BModel Type=Open-source2026.06 | 16 | 16.5 | |
| Llama3.2-11B-VisionModel Type=Open-source2026.06 | 15.5 | 26.5 | |
| GLM-4.6V-FlashModel Type=Open-source2026.06 | 10 | 18.5 | |
| InternVL3.5-4BModel Type=Open-source2026.06 | 5 | 15.5 |