AI Potluck
Back to Gap Map Model components / Scientific AI models

Orb

Orbital Materials
open weights / Overall score: 2.0

Orb is Orbital Materials' family of pretrained interatomic potentials for atomic simulations, covering inorganic crystals through the orb-v2 and orb-v3 series and molecular systems through OrbMol. The models predict energies, forces and stresses, expose a per-atom confidence head, integrate with ASE and TorchSim, and can be finetuned on a custom dataset with the packaged script.

Checkpoints are loaded through the orb_models.forcefield.pretrained module rather than declared as a Hub artifact. Verified 2026-09-10 via the GitHub API, the repository README, MODELS.md and LICENSE body.

Openness

3 high confidence
3.0
weights
open(Apache-2.0 ungated checkpoints loaded through the orb_models.forcefield.pretrained module)
data
open(MPTrj, Alexandria, OMat24, OMol25, and OPoly26 training datasets publicly released)
code
partial(inference, simulation wrappers, and a finetuning script
license
Apache-2.0(OSI)

Orb names a released public dataset behind every checkpoint and licenses the models under Apache-2.0, but the repository ships adaptation rather than a pretraining run, which keeps it off the top rungs.

Adoption

2 high confidence
2.0

Re-banded off stars. orb-models is a declared artifact, and sources/signal_routing.yaml sets the precedence pypi > huggingface > stars, so an authoritative usage route exists and the record may not fall through to a star count. 12,043 downloads in the trailing 30 days puts it at level 2, one above the 616 stars it used to carry.

Capability

2 high confidence
2.0

One attribute. Periodic and non-periodic support is coverage rather than a second scale, the direct and conservative variants are separate checkpoints, and the models predict energies, forces and stresses rather than generating structures.

Verified 2026-09-10