Recent developments in atomistic modeling: machine learning models and datasets, methods, software releases, and scientific events

Abstract

In this review, we summarize a remarkable series of developments in atomistic modeling that unfolded over just six weeks, from mid-May to the end of June 2025. This extraordinary sequence began on May 13, when Meta’s Fundamental AI Research team released the OMol25 dataset, a large-scale, high-accuracy quantum chemistry dataset, along with an accompanying paper on arXiv and the model on Hugging Face. On the same day, they also released universal models for atoms (UMA) on Hugging Face, with the related paper published on arXiv on June 30, 2025. Around the same period, the widely used package for atomistic calculations, ORCA, was updated to version 6.1 and officially released on June 17, introducing substantial methodological and performance improvements. In the area of semiempirical methods, the group of Prof. Grimme released a new and powerful method, g-xTB, on June 24. As the successor to the GFN family, g-xTB is a general-purpose extended tight-binding method that offers significantly improved accuracy at a modest computational cost. These advances coincided with WATOC 2025, the triennial meeting of the theoretical and computational chemistry community, held in Oslo from June 22 to 27, with g-xTB presented in session on June 26. Taken together, these developments represent one of the most dynamic and impactful periods in the recent history of atomistic modeling. During this time, we also released a new version of our platform, Atomistica.online 2025, our contribution aimed at enhancing accessibility and usability in molecular modeling.