MIT researchers reported in October 2026 in Nature Communications a multimodal AI model that combines three-dimensional atomic structures with natural-language descriptions drawn from experimental papers and first-principles computational databases. The descriptions cover chemical composition and structure as well as properties, processing considerations and potential applications. The paper says the system identified better material designs than models trained on text alone.

In one example, when asked to find materials suited for solid-state lithium-ion battery electrolytes, the model's top results included materials independently identified as promising candidates; researchers also said it learned geometric representations text-only models did not. Ju Li, a professor in MIT's departments of Materials Science and Engineering and Nuclear Science and Engineering, said structure is central to materials science and that the aim was to connect atomic structure with properties and processing.

The work was supported in part by the U.S. Defense Threat Reduction Agency, the U.S. Defense Advanced Research Projects Agency and the National Research Foundation of Korea. The researchers said they hope other materials scientists will use the approach, while noting existing literature is skewed toward a few popular crystal classes.