Galactica: A Large Language Model for Science

by Meta AI · free · Last verified 2026-03-17

Galactica is a series of LLMs (125M to 120B parameters) trained on a curated corpus of 48M scientific papers, reference materials, and knowledge bases, designed to store, combine, and reason over scientific knowledge. Despite controversy at release, Galactica established important design principles for scientific LLMs including citation token formatting and working memory prompting.