https://english.news.cn/20260604/bcc263ea784243dc8c686591cd10ee7e/c.html

A joint research team from Chinese institutions has upgraded a new-generation large atomic model DPA4. It is capable of simulating energy variations and force interactions between atoms in the fields such as inorganic crystals and small organic molecules. Its application covers a wide range of research areas, including catalysts, battery materials, semiconductors, and drug molecules, according to the report.

The model is powered by artificial intelligence (AI) and has achieved four generations of upgrades in four years. It tops Matbench Discovery, an influential dynamic benchmark in the field of AI-driven inorganic materials discovery. The core breakthrough of DPA4 is the ability to achieve high accuracy at a very low computational cost.

This latest large atomic model, developed by the researchers from AI for Science Institute (AISI) in Beijing, DP Technology, Peking University and Institute of Applied Physics and Computational Mathematics, has been made available as an early-access version in open-source communities for researchers to test and experience. The team has said that an official open-source version will be released at a later date.

As one of the foundational infrastructures in the field of scientific intelligence, the DPA model is undergoing continuous iteration, transforming large-scale and high-throughput atomic simulations from a “luxury” into a “daily necessity”.