Benchmark RadarChildEval
ChildEval is a benchmark for evaluating LLMs' ability to infer and follow child-centered preferences in long-context conversations. It contains 29K synthesized persona profiles of children aged 3-6, with explicit and implicit preference expressions across five top-level and fourteen sub-level categories.
- Released
- 2026-05-27
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Personalization for children is under-explored relative to adults. ChildEval provides a protocol to test whether LLMs can infer and follow child-specific preferences, addressing a gap in personalized conversational AI evaluation.
Motivation
While LLMs enable personalized chatbots, their effectiveness in child-centered personalization remains unclear, as systematic evaluation of child-specific preferences is still lacking.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.