List your company

Safety Classification on SafeEditBench

100Policy L1 Success RateLlama Guard
-4235077Mar 1, 2026

Evaluation results

MethodLinks
Llama Guard2026.031000042.37022.73
Qwen2-VL-7B2026.0365.2515.382460.874436.65
SafeGuard-VLTraining=RL+SafeEditTrain2026.0354.2423.0835.972.5866.1749.43
LlavaGuard-7B2026.0349.1516.6731.8266.1364.6244.16
Qwen2.5-VL-7B2026.0347.4620.5937.3670.8770.3448.68
ShieldGemma22026.0344.925.6314.1237.2951.0627.5
SafeGuard-VLTraining=RL2026.0344.0720.2935.1670.464.7145.59
QwenGuard-7B2026.0316.113.5632.6166.0252.3432.76
LlavaGuard-0.5B2026.0311.029.6827.9449.3749.730.52
GLM-4V-9B2026.03014.6827.5467.9573.3741.16
QwenGuard-3B2026.03001.6322.3905.19