orbmol-v2

OrbMol-v2

OrbMol-v2 extends the OrbMol architecture (a continuation of the orb-v3 series, trained on Open Molecules 2025 (OMol25) and OPoly26 — over 100M ωB97M-V/def2-TZVPD calculations) with learnable per-atom electrostatics:

• LatentChargeHead predicts a per-atom latent charge feature, constrained to sum to the system total charge.
• CoulombModule adds a long-range Coulomb energy on top of the GNN — direct bare-1/r Coulomb sum for non-periodic systems, Particle Mesh Ewald via nvalchemiops for periodic systems.
• ChargeConditionedEnergyHead is the energy head, conditioned on the latent per-atom charges.
• ChargeSpinConditioner mixes the system's net charge and spin (multiplicity) into the backbone as system-level features. The model uses spin directly as a conditioning input — there is no per-atom spin prediction in this release.

Caution: While the model does predict per-atom charge values as latent features in the charge head, the model has not seen any per-atom charge values during training — these are emergent from optimisation against energies and forces alone. They should therefore be treated with caution: while in at least some cases they appear to correspond to the correct physical values, the reliability and generality of this correspondence is unclear and is the subject of ongoing investigations.

Usage

Install orb-models (≥ the version that introduces orbmol-v2) and load the model:

from orb_models.forcefield.pretrained import orbmol_v2

model, atoms_adapter = orbmol_v2(device="cuda")

# atoms.info["charge"] (net charge) and atoms.info["spin"] (multiplicity, = 2S+1)
# must be set before passing atoms through the model.

Energies are returned in fp64 by default to preserve kJ/mol resolution against OMol25-scale references; pass fp64_energy=False to model.predict(...) to opt out.

Files

- orbmol-v2-teqabfhg-20260523.ckpt — flat PyTorch state_dict (~100 MB), with EMA shadow params and the persistent coulomb_constant buffer baked in.

Training data

OMol25 and OPoly26 — ωB97M-V/def2-TZVPD on diverse molecular systems including metal complexes, biomolecules, and electrolytes.

License

Apache-2.0.