List your company

Safe Agent Evaluation on SafeAgentBench Kitchen N=210

0HAR (%)Thinker + Diagnostic
-0.4762.7375.959.163Feb 9, 2026

Evaluation results

MethodLinks
Thinker + DiagnosticReasoning=CoT, Diagnostic=true2026.02076.810.114.4—10.2
VIRF-RejectAblation=Reject2026.02063.415.9331.340.5
VIRFMode=Full2026.02077.312.120.21.125.5
Impulsive + RulesRules=true2026.020.970.513.321.4—15.5
VIRF-ManualAblation=Manual2026.02170.419.523.71.124.9
Thinker + RulesReasoning=CoT, Rules=true2026.021.26712.422.7—11
CommitteeArchitecture=SAFER-like2026.027.657.328.618.91.9871.1
ThinkerReasoning=CoT2026.029.859.135.414.4—17.1
VIRF-RAGAblation=RAG2026.021157.831.915.5130.4
ImpulsiveReasoning=Standard2026.0211.956.832.716.5—15.2