KIDBench (No Cue) - Boundary-setting: leaderboard

KIDBench No Cue Boundary-setting rates whether an LLM answer to a child aged 7 to 11 sets age-appropriate limits around unsafe, private or adult topics and points the child to a safe adult when needed, when the prompts give no sign that the user is a child (KIDBench, 2026). A DeepSeek-V4-Pro judge scores each answer from 1 to 5, averaged over ten safety-relevant categories; higher is better.

Metric: Boundary-setting score (1-5): whether the response sets age-appropriate limits around unsafe, private or adult topics and points the child to a safe adult when needed, rated 1-5 by a DeepSeek-V4-Pro judge with the KIDBench rubric and category-specific rules, temperature 0 for responder and judge, averaged on the KIDBench child-safety prompts for children aged 7 to 11 (50 per category, 10 categories), single-turn, no-cue prompts (no sign that the user is a child); higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1DeepSeek V4 Flash2.97#267
2Qwen 3.6 27B2.87#351
3Llama 3.3 70B Instruct2.82#1385
4Gemma 3 12B2.8#2059
5GPT-5 Mini2.79#364
6Qwen 3.5 4B2.77#1241
7Gemini 3.1 Flash Lite2.71#371
8Claude Haiku 4.52.69#583
9Gemma 4 31B (IT)2.66#350
10Qwen 3 8B2.6#1873
11Gemma 3 4B2.52#5339
12KIDBench Llama-3.2-3B (checkpoint unspecified)2.46
13Llama 3.1 8B Instruct2.45#4600

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=kidbench-no-cue-boundary-setting · How It Works · Data refreshed daily, snapshot 2026-11-21.