HLE Dataset 多学科知识数据集

HLE 全称为 Humanity’s Last Exam，是一个多模态基准测试，旨在涵盖广泛学科领域的前沿知识，由 AI 安全中心 (Center for AI Safety) 、 Scale AI 于 2025 年发布，相关论文成果为「Humanity’s Last Exam」。该数据集由全球各学科专家共同开发，包含 3k 个问题，覆盖数十个学科，包括数学、人文学科和自然科学等领域的多项选择题和简答题，适合自动化评分。

HLE, short for Humanity’s Last Exam, is a multimodal benchmark designed to cover cutting-edge knowledge across a wide range of academic disciplines. Released in 2025 by the Center for AI Safety and Scale AI, its corresponding academic paper is titled *Humanity’s Last Exam*. Developed collaboratively by experts from diverse disciplines worldwide, this dataset contains 3,000 questions spanning dozens of fields, including multiple-choice questions and short-answer questions from domains such as mathematics, humanities, natural sciences and other areas, and supports automated scoring.