Large Language Models (LLMs) for Materials Design

Abstract In recent years, rapid advances in large language models (LLMs) have been witnessed, while materials scientists have quickly adapted to exploit their potential. This review surveys the latest developments at the intersection of LLMs and materials science. Both general‐purpose and materials‐specific LLMs, as well as their theoretical and technical backgrounds, are first discussed. Their core capabilities are then detailed, including domain‐specific question answering for materials design, automated data extraction, semantic‐driven material design, synthesis planning, property prediction, and emphasize on inverse materials design as well as integration with materials simulation (e.g., density function theory and crystal structure prediction) workflows. These capabilities are illustrated through case studies in metals/alloys, metal organic frameworks, glasses/ceramics, photovoltaics, catalysis, and batteries, followed by a discussion of advances in retrieval‐augmented generation (RAG) and agent systems for materials science. Finally, current challenges and promising future directions of LLMs for materials science are outlined, especially physics‐informed transformer (PIT) or materials‐informed transformer (MIT) architecture that may effectively bridge the gap between materials science and LLMs (i.e., “LLMs for materials” and “materials for LLMs”).