KIDBench - Cultural Alignment (China): leaderboard
Metric: Cultural Alignment score (1-5): whether the response adapts guidance, help-seeking pathways and social expectations to China, rated by a DeepSeek-V4-Pro judge with country-specific cultural rules (temperature 0), on the 500 implicit-cue KIDBench prompts in English with a system instruction naming a child aged 7 to 11 in China, mean over the 10 categories, single-turn; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.6 27B | 4.24 |
| 2 | Gemma 3 12B | 3.97 |
| 3 | GPT-5 Mini | 3.83 |
| 4 | Gemini 3.1 Flash Lite | 3.82 |
| 5 | DeepSeek V4 Flash | 3.73 |
| 6 | Gemma 4 31B (IT) | 3.72 |
| 7 | Claude Haiku 4.5 | 3.51 |
| 8 | Qwen 3.5 4B | 3.45 |
| 9 | Llama 3.3 70B Instruct | 3.44 |
| 10 | Qwen 3 8B | 3.43 |
| 11 | Gemma 3 4B | 3.28 |
| 12 | Llama 3.1 8B Instruct | 3.26 |
Interactive version: theaggregate.ai/benchmark?slug=kidbench-cultural-alignment-china · How It Works · Data refreshed daily, snapshot 2026-10-07.