Multi-Modal Foundation · Orbital-Materials
OrbMol
A molecular foundation-model interatomic potential predicting energies, forces, and properties across chemical systems.
What OrbMol is
OrbMol is a universal interatomic potential developed by Orbital-Materials for molecular systems, spanning organic and metal-organic chemistry. It addresses the computational bottleneck of high-level quantum mechanical calculations by predicting atomic energies, forces, and properties to support catalyst discovery and pharmaceutical workflows.
Unlike prior iterations of the model family, OrbMol accepts system-level total charge and spin multiplicity as explicit inputs. This allows it to handle open-shell configurations, ionic electrolytes, biomolecules, and transition metal complexes that conventional potentials often fail to represent accurately.
How it works
OrbMol is built upon the Orb-v3 neural network architecture. It functions as an interatomic potential operating over atomic graphs with a 6 Å radius cutoff and an upper bound of 120 neighbours per atom during neighbour list computation.
The model was trained on the Open Molecules 2025 (OMol25) dataset, comprising more than 100 million density functional theory (DFT) calculations performed at the ωB97M-V/def2-TZVPD level of theory. Two checkpoints are available: a default conservative variant where forces correspond directly to energy gradients, and a direct variant (OrbMol-direct) where forces are predicted directly rather than as exact gradients.
Key facts
- Architecture
- Orb-v3
- Training data
- Open Molecules 2025 (OMol25; over 100M DFT calculations at ωB97M-V/def2-TZVPD)
- Cutoff radius
- 6 Å
- Maximum neighbours
- 120
- Checkpoints
- OrbMol (conservative) and OrbMol-direct
What you can use it for
- Open-Shell and Radical Chemistry
- Simulating radical intermediates and open-shell molecular systems by explicitly conditioning predictions on spin multiplicity.
- Electrolyte and Ionic System Modelling
- Calculating forces and energies for charged ionic electrolytes through direct input of the system's total net charge.
- Metal Complex and Catalyst Workflows
- Evaluating metal-organic complexes and organometallic catalytic species across varying oxidation and spin states.
- Biomolecular Simulation
- Predicting atomic forces and molecular properties within biomolecular systems in pharmaceutical research pipelines.
Run it here
We list this model but cannot run it yet: no pipeline_tag — needs a human.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Explicitly accepts total charge and spin multiplicity to capture charged and open-shell states.
- Trained on a extensive dataset of over 100 million high-level ωB97M-V/def2-TZVPD DFT calculations.
- Provides both a conservative formulation ensuring energy-conserving forces and a direct force prediction variant.
Watch out for
- Interactions beyond the 6 Å cutoff radius or exceeding the 120-neighbour maximum are not captured by the neighbour list.
- Forces in the OrbMol-direct variant are not exact gradients of the energy surface.
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 · apache-2.0
The apache-2.0 licence carries terms worth reading before you depend on it.
Questions
Can I use OrbMol commercially?
Check the licence before commercial use. The apache-2.0 licence carries terms worth reading before you depend on it.
What is the difference between OrbMol and OrbMol-direct?
The standard OrbMol checkpoint is conservative, meaning forces are the exact gradient of the predicted energy. In the OrbMol-direct checkpoint, forces are predicted directly and are not mathematically guaranteed to be conservative gradients.
What additional inputs does OrbMol require compared to earlier Orb versions?
OrbMol requires the specification of total charge and spin multiplicity alongside atomic structures, enabling the model to distinguish different electronic and ionisation states.
What quantum mechanical method was used for the training data?
The model was trained on the OMol25 dataset, which used the ωB97M-V density functional with the def2-TZVPD basis set.
How to cite it
Benjamin Rhodes, Sander Vandenhaute, Vaidotas Šimkus, James Gin, Jonathan Godwin, Tim Duignan, Mark Neumann. Orb-v3: atomistic simulation at scale. arXiv:2504.06231, 2025.
Written from OrbMol'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.