List your company

Agentic Safety Moderation on CIK-Bench

33ASRw/o guardrail
13.2418.3723.528.63May 28, 2026

Evaluation results

MethodLinks
w/o guardrailGuardrail=None2026.0533——————
Qwen3Guard-Gen-4BGuardrail=Qwen3Guard-Gen-4B2026.0532-2.860.06360.06760.01290.013313.98
AgentDoG 1.5-0.8BBase Model=Qwen3.5 Thinking Base, Size=0.8B2026.0530-8.570.12680.13370.01470.0148305.42
AgentDoG 1.5-4BBase Model=Qwen3.5 Thinking Base, Size=4B2026.0515-51.430.16710.17960.02060.0207377.18
Llama-Guard-3-8BGuardrail=Llama-Guard-3-8B2026.0514-54.290.0290.03170.01150.01194.14