*This dataset supports online use.Click here to jump.

The s1K dataset is a high-quality reasoning dataset released by Fei-Fei Li's team in 2025. It contains 1k questions and their detailed reasoning trajectories and answers. These answers are derived from the distillation results of Google's Gemini Thinking Experimental. The dataset covers 50 different fields, including probability theory, quantitative interview questions, and Olympic questions, ensuring that the model can handle various types of reasoning tasks. The related paper results are "s1: Simple test-time scaling".

The dataset is designed to enable efficient model fine-tuning through minimal data engineering and demonstrate excellent performance in inference tasks.