Meta unveiled a groundbreaking initiative on Wednesday that promises to transform scientific research, particularly in computational chemistry. The tech giant released Open Molecules 2025 (OMol25), an unprecedented dataset containing more than 100 million molecular simulations, alongside a powerful new AI model designed to accelerate scientific discoveries in fields ranging from drug development to materials science.

Unprecedented Scale of Computation

The sheer scale of Meta’s effort is staggering. The project consumed 6 billion compute hours in Meta’s global data centers—approximately ten times more than any previous scientific dataset.

“We’re talking about two orders of magnitude more compute than any kind of academic data set that’s ever been made,” said Sam Blau, a research scientist at Lawrence Berkeley National Laboratory who collaborated with Meta. “It’s going to dramatically change how people do computational chemistry.”

To put this computational demand in perspective, as Blau told Mirage News, “it would take you over 50 years to run these calculations with 1,000 typical laptops.”

Meta’s Fundamental AI Research (FAIR) team leveraged the company’s vast global network of computing resources, strategically utilizing periods of spare bandwidth when users in different parts of the world weren’t actively browsing Instagram or Facebook.

Quantum Leap for Molecular Modeling

What makes OMol25 revolutionary is both its size and complexity. Previous molecular datasets typically contained simulations with 20-30 atoms on average and covered only a handful of elements. In contrast, OMol25’s configurations are ten times larger, with up to 350 atoms from across most of the periodic table, including heavy elements and metals that are notoriously difficult to simulate accurately.

The dataset captures an extensive range of molecular interactions and dynamics involving both organic and inorganic molecules. Meta employed Density Functional Theory (DFT)—complex mathematical calculations that map how atoms physically interact—to create this comprehensive repository.

Accelerating Scientific Discovery

Alongside the dataset, Meta released a new family of AI models called Universal Model for Atoms (UMA), which can dramatically speed up scientific research. These models can perform complex chemistry calculations up to 10,000 times faster than traditional methods.

The implications for scientific advancement are profound. Researchers can now explore new possibilities for drug development, battery technology, sustainable materials, and other innovations at unprecedented speeds.

“Scientists have long created mathematical representations of the way atoms and molecules interact, but it has been prohibitively expensive to do those calculations on a large scale,” Semafor reported. Meta’s initiative removes this barrier, democratizing access to high-quality molecular data.

Beyond Chemistry: Advancing General AI

Meta’s investment in this project extends beyond pure scientific altruism. According to Semafor, the FAIR team believes that the new dataset and AI model will help advance overall AI technology toward what they call “Advanced Machine Intelligence.”

To achieve this goal, AI systems must build a “world model”—a fundamental understanding of the physical world around them. Since atoms are the building blocks of matter, understanding their interactions is a crucial step toward more sophisticated AI.

This initiative follows Meta’s previous efforts in open science, including Meta Open Catalyst, Meta Open DAC, and Meta Open Materials. Unlike many industry-led efforts that remain proprietary, Meta has consistently released comprehensive datasets and state-of-the-art models to the broader academic community.

The Open Molecules 2025 dataset and UMA models are now available to the scientific community, potentially accelerating breakthroughs in multiple fields while simultaneously advancing the frontier of artificial intelligence research.