Multi-Modal Foundation · facebook
OMol25
Predicts molecular properties and energies for chemistry and materials science researchers.
What OMol25 is
OMol25 (Open Molecules 2025) is a collection of baseline machine learning models and benchmark datasets developed by Facebook for molecular property prediction across chemistry, materials science, and engineering.
The models address the computational cost of electronic structure calculations by learning quantum chemical properties from density functional theory (DFT) data. The release provides checkpoints from multiple model architectures designed for both direct prediction and energy-conserving property estimation across molecular and polymer domains.
How it works
OMol25 provides pre-trained checkpoints based on two model architectures: eSEN (available in small and medium configurations) and AllScAIP (available in medium configurations). Models are trained on the full OMol25 dataset and are formulated either as direct property predictors or as energy-conserving variants.
The models interface with molecular simulation workflows through a custom Atomic Simulation Environment (ASE) calculator within the fairchem library. Training data is generated with specific DFT parameters that match companion resources, including the OMol Closed Shell Hamiltonian (OMol_CSH) dataset and the polymer-focused Open Polymers 2026 (OPoly26) dataset.
Key facts
- Release year
- 2025
- Category
- Molecular property prediction
- Architectures
- eSEN, AllScAIP
- Training data
- OMol25
- Simulation interface
- fairchem ASE-calculator
What you can use it for
- Molecular Property Prediction
- Estimates molecular electronic and chemical properties directly from molecular geometry without running full quantum chemistry simulations.
- Atomistic Simulations
- Drives molecular dynamics and geometry relaxations by plugging checkpoints into the Atomic Simulation Environment (ASE) via fairchem.
- Energy-Conserving Dynamics
- Performs molecular dynamics trajectories using dedicated conserving checkpoints (eSEN or AllScAIP) that adhere to energy conservation laws.
- Polymer Modelling
- Evaluates macromolecular systems in conjunction with the OPoly26 dataset, which shares identical DFT parameters with OMol25.
Run it here
We list this model but cannot run it yet: gated on the Hub — never run past a gate the user did not accept.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Provides multiple architecture scales, including small and medium checkpoints for both eSEN and AllScAIP.
- Supplies both direct and energy-conserving checkpoint formulations for different simulation requirements.
- Integrates directly into standard computational chemistry workflows using an ASE calculator in fairchem.
- Uses density functional theory parameters identical to the OPoly26 dataset, avoiding inconsistencies across molecular and polymer tasks.
Watch out for
- The large model configuration (eSEN-lg-direct) is not available in the initial checkpoint release.
- Model predictions reflect the specific density functional theory calculation settings used to generate the OMol25 training set.
- Execution depends on the fairchem library and its custom calculator environment.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · other
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use OMol25 commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Which model architectures are available in OMol25?
OMol25 provides checkpoints for the eSEN and AllScAIP architectures. These include eSEN-sm-direct, eSEN-sm-conserving, eSEN-md-direct, AllScAIP-md-direct, and AllScAIP-md-conserving.
How are the OMol25 checkpoints run in simulation workflows?
Checkpoints can be executed in Python through a custom Atomic Simulation Environment (ASE) calculator provided by the fairchem library.
What is the relationship between OMol25 and OPoly26?
OPoly26 is an extension of OMol25 focused specifically on polymers. Because both datasets share identical DFT settings, they can be combined without methodology mismatches.
What is the difference between direct and conserving checkpoints?
Direct checkpoints predict output properties directly, whereas conserving checkpoints are designed to conserve energy throughout calculations and simulations.
How to cite it
Daniel S. Levine, Muhammed Shuaibi, Evan Walter Clark Spotte-Smith, Michael G. Taylor, Muhammad R. Hasyim, Kyle Michel, Ilyes Batatia, Gábor Csányi, Misko Dzamba, Peter Eastman, Nathan C. Frey, Xiang Fu, Vahe Gharakhanyan, Aditi S. Krishnapriyan, Joshua A. Rackers, Sanjeev Raja, Ammar Rizvi, Andrew S. Rosen, Zachary Ulissi, Santiago Vargas, C. Lawrence Zitnick, Samuel M. Blau, Brandon M. Wood. The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models. arXiv:2505.08762 [physics.chem-ph], 2025.
Written from OMol25's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.