Meta’s Fundamental AI Research (FAIR) has compiled over 100 million chemistry models simulations using over 6 billion compute hours to create the Open Molecules 2025 (OMol25¹) machine learning dataset along with an AI agent, UMA (Universal Model for Atoms AI).

The new dataset with AI agent is said to be a way to speed up the creation of new drugs and other chemistry challenges like batteries . The problem revolves around the costly calculations on a large scale scientist were facing for creating mathematical representations of the way atoms and molecules interact. Meta, has been using unused data center compute capacity to create the Open Molecules dataset using complicated maths known as the Density Functional Theory². The new AI, UMA, can get the same results as the traditional methods, but at a rate of 10,000 times faster.

The Open Molecules 2025 dataset is based on four broads areas: small molecules, biomolecules, metal complexes, and electrolytes.

1. https://arxiv.org/abs/2505.08762 ↩︎
2. https://en.wikipedia.org/wiki/Density_functional_theory ↩︎