Prompt Injection Detection on EIA
97.39RecallGPT-4.1
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4.1Model Type=Closed-source APIs2026.04 | 97.39 | |
| WebAgentGuard-4BModel Type=Ours2026.04 | 95.69 | |
| WebAgentGuard-8BModel Type=Ours2026.04 | 93.71 | |
| GPT-4oModel Type=Closed-source APIs2026.04 | 93.07 | |
| Prompt-Guard-2-86MModel Type=Guard models2026.04 | 92.79 | |
| Qwen3-VL-Instruct-4BModel Type=Open-source instructed models2026.04 | 78.45 | |
| GPT-4o-MiniModel Type=Closed-source APIs2026.04 | 63.18 | |
| GuardReasoner-VL-7BModel Type=Guard models2026.04 | 58.87 | |
| Qwen3-VL-Instruct-8BModel Type=Open-source instructed models2026.04 | 44.88 | |
| Llama-Guard-3-Vision-11BModel Type=Guard models2026.04 | 21.77 | |
| Llama-3.2-Vision-Instruct-11BModel Type=Open-source instructed models2026.04 | 19.93 | |
| Prompt-Guard-1-86MModel Type=Guard models2026.04 | 10.81 | |
| Qwen2.5-VL-Instruct-7BModel Type=Open-source instructed models2026.04 | 8.27 |