List your company

Harmful-input detection on Safety Evaluation Suite (AEG, AEG2, HB-A, HB-V, OAI, SAL, SST, Toxic, WGT, WJB)

89.8AEG ScoreWildGuard
65.8872.0978.384.51Jun 26, 2026

Evaluation results

MethodLinks
WildGuard2026.0689.881.692.299.172.79799.570.788.797.888
Robust Harmful Features detectorBackbone=Gemma-9B2026.068880.793.198.568.994.199.555.580.183.781.6
Robust Harmful Features detectorBackbone=Gemma-2B2026.0687.88191.699.168.897.69964.58288.685.3
Robust Harmful Features detectorBackbone=Qwen-14B2026.0687.581.492.999.465.298.610060.880.288.484.7
Robust Harmful Features detectorBackbone=Qwen-7B2026.0686.582.590.198.769.898.599.566.677.990.786
Robust Harmful Features detectorBackbone=Llama-2-7B2026.0686.282.279.79976.164.99978.8828875.2
Robust Harmful Features detectorBackbone=Gemma-27B2026.0685.477.392.598.564.896.697.143.781.285.479.8
Robust Harmful Features detectorBackbone=Llama-3-8B2026.0682.483.989.197.476.396.499.580.48294.388.8
VanillaBackbone=Llama-2-7B2026.0677.876.983.298.168.658.195.954.380.188.566.4
Qwen3Guard2026.067682.783.698.780.893.297.482.486.791.287.7
VanillaBackbone=Llama-3-8B2026.0675.778.386.494.173.788.79858.479.988.879.7
VanillaBackbone=Qwen-7B2026.0674.174.156.589.654.343.788.540.461.448.350.6
VanillaBackbone=Qwen-14B2026.0672.873.466.686.455.132.288.936.166.860.647
LlamaGuard32026.0671.777.184.698.17945.599.553.976.867.959.9
VanillaBackbone=Gemma-9B2026.0669.272.577.197.251.639.394.942.97257.451.4
VanillaBackbone=Gemma-27B2026.0668.47176.697.553.841.593.842.372.655.251.9
LlamaGuard42026.0667.871.585.295.773.566.498.551.573.869.666.6
VanillaBackbone=Gemma-2B2026.0666.871.473.797.857.32791.448.369.851.547