FAIRChem is the open-source library from Meta’s FAIR Chemistry team that collects its data, models, and applications for machine learning in materials science and quantum chemistry. The GitHub repository has around 2,300 stars and remains actively developed in 2026, serving as the home of widely used models for catalysis and materials discovery.
FAIRChem bundles the models behind the Open Catalyst and Open Materials projects into one Python library (fairchem-core). Its flagship UMA (Universal Model for Atoms) family provides pretrained universal interatomic potentials that can predict energies and forces for molecules, surfaces, and crystals, enabling fast simulation-driven screening without running expensive quantum chemistry calculations for every candidate.
The datasets behind FAIRChem are among the largest in computational chemistry: the Open Catalyst project alone contains well over a million adsorption relaxations computed with density functional theory. By training on that scale of data, FAIRChem’s models aim to approximate DFT accuracy at a fraction of the cost, and the library doubles as the reference implementation for evaluating new models on these community benchmarks.
FAIRChem is distributed as the fairchem-core package on PyPI and supports both single-GPU experimentation and multi-GPU training for researchers who want to train potentials from scratch. The team maintains demos that run in the browser and in Codespaces, lowering the barrier for chemists who are new to machine learning to try state-of-the-art models on their own structures.
FAIRChem is free and open source. The library, pretrained models, and datasets are publicly available on GitHub and PyPI at no cost.
Computational chemists, materials scientists, and ML researchers who want ready-made machine learning potentials and datasets for catalysis and materials discovery.