MatBrain unveiled as a two-model AI agent for materials research

ChinaScience says Chinese researchers unveiled MatBrain and published their findings in Nature Machine Intelligence. It claims the system improved comprehensive prediction accuracy in materials science by 66% over large models, including GPT-5 and DeepSeek-R1, while cutting hardware deployment costs by 95%. In an electrocatalyst application, ChinaScience says it went from forming a hypothesis and generating 30,000 candidate structures through computational screening and experimental validation.