KIDBench (No Cue) - Hate, Bias, and Identity Attacks: leaderboard
Metric: Category score (1-5): mean of the five child-safety dimensions, each rated 1-5 by a DeepSeek-V4-Pro judge with the KIDBench rubric and category-specific rules, temperature 0 for responder and judge, over the 50 Hate, Bias, and Identity Attacks prompts of KIDBench for children aged 7 to 11, single-turn, no-cue prompts (no sign that the user is a child); higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V4 Flash | 3.62 |
| 2 | Llama 3.3 70B Instruct | 3.52 |
| 3 | GPT-5 Mini | 3.3 |
| 4 | Gemma 3 12B | 3.18 |
| 5 | Qwen 3.6 27B | 3.05 |
| 6 | Gemini 3.1 Flash Lite | 3.01 |
| 7 | Qwen 3.5 4B | 3 |
| 8 | Qwen 3 8B | 2.98 |
| 9 | Claude Haiku 4.5 | 2.97 |
| 10 | Llama 3.1 8B Instruct | 2.96 |
| 11 | Gemma 3 4B | 2.93 |
| 12 | Gemma 4 31B (IT) | 2.86 |
Interactive version: theaggregate.ai/benchmark?slug=kidbench-no-cue-hate-bias-and-identity-attacks · How It Works · Data refreshed daily, snapshot 2026-10-07.