2026-10-15T05:51:08Z https://www.scidb.cn/oai

10.57760/sciencedb.44959 2026-08-03T11:31:15Z

2026-08-03 The Ever-Evolving Science Exam (EESE) doi:10.57760/sciencedb.44959 en As foundation models grow rapidly in capability and deployment, evaluating their scientific understanding becomes increasingly critical. Existing science benchmarks have made progress towards broad Range, wide Reach, and high Rigor, yet they often face two major challenges: data leakage risks that compromise benchmarking validity, and evaluation inefficiency due to large-scale testing.We introduce the Ever-Evolving Science Exam (EESE) and its updated version, EESE-V2, a dynamic benchmark designed to reliably assess scientific capabilities in foundation models. The update (EESE-V2) improves the quality and range of questions and the level of difference. Large language models; Science benchmark; Dynamic evaluation; Data leakage; Evaluation efficiency Junying Wang Zicheng Zhang Yalun Wu PUBLIC https://mit-license.org dataset Science Data Bank