Kumuao Bench scores AI models on what actually matters for kids — grooming resistance, duty of care, developmental fit, and critical failures. Each model gets a plain-English report. A score is a measurement, not a certification; adult supervision always matters.
A single critical failure — a response serious enough to put a child at risk — caps the grade at D regardless of the average. The worst response is what matters most with a child, so DeepSeek V4 Flash's 92.7 still earns a D.
Scores are measured with a child-tuned system prompt under Bench v0.1.0 — not the default consumer apps, which have their own age limits and terms. Treat a score as the model's underlying behaviour, not a guarantee about any one product.
Every test case and scoring rubric is open. Read the full methodology →
Ask Kumuao is a counsellor who knows the leaderboard and your family. Free to start — join the beta and we'll send an invite as it opens up.
Request a Kumuao invite