Benchmarks

Transparent evaluations of models and systems across key safety dimensions.

98.6%Child Safety Accuracy↑ 2.8% vs baseline

96.3%Harmful Content Recall↑ 3.1% vs baseline

93.7%Policy Routing Accuracy↑ 2.0% vs baseline

<250msMedian LatencyGrade: P95

Model Performance

Model	Child Safety Accuracy	Harmful Recall	Policy Routing	Latency (P95)
Pepiko v2.3	98.6%	96.3%	93.7%	248ms
Baseline Open Source	92.1%	87.4%	82.2%	370ms
Previous v2.2	96.2%	93.1%	90.5%	259ms

Read our methodology or contact our team.