Skip to content

OMat24

Pre-trained checkpoints for material property prediction

Visit official site

What OMat24 is

The release is described as a set of EquiformerV2 checkpoints from Meta FAIR Chemistry for predicting material properties with machine learning.

Model notes

  • Sizes of 31M, 86M and 153M are offered
  • Checkpoints are trained on OMat, MPtrj and sAlexandria
  • Results are shown on Matbench Discovery

Application scenarios

  • Researchers predict thermodynamic properties
  • Chemists optimise material synthesis paths
  • Teams predict strength and durability for experiments

Pros & cons

✓ What we like

  • Pre-trained checkpoints can be used directly
  • A custom ASE calculator works in the fairchem library

! What to watch out for

  • Cell relaxation needs a specific branch
  • Setup requires the fairchem library and documentation

FAQ

Which sizes are listed?

The material names 31M, 86M and 153M versions.

Which architecture is used?

The checkpoints use the EquiformerV2 architecture.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review