List your company

Prompt Injection Detection on VPI-Bench

87.58RecallWebAgentGuard-8B
-3.503220.143443.7967.4366Apr 14, 2026

Evaluation results

MethodLinks
WebAgentGuard-8BModel Type=Ours2026.0487.58
WebAgentGuard-4BModel Type=Ours2026.0485.95
GPT-4oModel Type=Closed-source APIs2026.0478.43
GPT-4o-MiniModel Type=Closed-source APIs2026.0469.93
GPT-4.1Model Type=Closed-source APIs2026.0469.61
Llama-Guard-3-Vision-11BModel Type=Guard models2026.0468.95
Prompt-Guard-1-86MModel Type=Guard models2026.0464.38
Llama-3.2-Vision-Instruct-11BModel Type=Open-source instructed models2026.0443.79
Qwen3-VL-Instruct-8BModel Type=Open-source instructed models2026.0442.16
Qwen3-VL-Instruct-4BModel Type=Open-source instructed models2026.0440.52
Qwen2.5-VL-Instruct-7BModel Type=Open-source instructed models2026.0414.71
GuardReasoner-VL-7BModel Type=Guard models2026.044.9
Prompt-Guard-2-86MModel Type=Guard models2026.040