NICE - Observational Imitation: leaderboard

NICE (2026) is a theory-grounded diagnostic benchmark of social intelligence with Chinese-context items that ask the model to rank candidate responses from socially optimal to boundary-violating; an item counts only when the whole ranking is right. This board is ranking accuracy in percent on the observational imitation dimension; higher is better.

Metric: Ranking accuracy (%) on the observational imitation dimension on NICE (137 human-written items in Chinese social contexts, each a ranking of an optimal, a suboptimal and a boundary-violating response), an item counts only when the full ranking matches the gold order, first-party APIs at temperature 0, three runs pooled; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)72.2#99
2Claude Opus 4.769.4#85
3DeepSeek V4 Pro61.1#196
4Qwen 3.6 Plus61.1#238
5GPT-5.555.6#54

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

Interactive version: theaggregate.ai/benchmark?slug=nice-observational-imitation · How It Works · Data refreshed daily, snapshot 2026-11-21.