generated: '2026-08-26'
method: searched
source: https://github.com/orbital-materials/orb-models/releases + https://raw.githubusercontent.com/orbital-materials/orb-models/main/README.md
name: Orbital Materials — orb-models release history
scope: >-
  This is the release log for the `orb-models` open-source Python library and its model
  weights, NOT an API changelog. Orbital Materials publishes no HTTP API and therefore no
  API changelog; this is the only dated, first-party, machine-readable record of what the
  company ships and when.
scheme: semver
current_version: 0.7.0
current_version_date: '2026-05-26'
changelog_urls:
- https://github.com/orbital-materials/orb-models/releases
- https://github.com/orbital-materials/orb-models#updates
entries:
- version: v0.7.0
  date: '2026-05-26'
  breaking: true
  highlights:
  - Release of OrbMol-v2, adding a CoulombModule for long-range electrostatics on top of
    the OrbMol architecture (direct Coulomb summation for non-periodic systems, Particle
    Mesh Ewald via nvalchemiops for periodic systems); trained on OMol25 and OPoly26.
  - GSCDB138 Normalized Error Ratio drops from 6.05 to 1.62.
  - Full-model compilation — model.compile(...) now wraps the full regressor for all
    models, ~1.7x speedup at 10k atoms on a single 80 GB GPU.
  breaking_changes:
  - model.predict(...)["energy"] now returns fp64 by default to preserve kJ/mol
    resolution against OMol-scale references; pass fp64_energy=False to opt out.
  source: https://github.com/orbital-materials/orb-models/releases/tag/v0.7.0
- version: v0.6.2
  date: '2026-03-18'
  breaking: null
  highlights: []
  note: Release notes not captured beyond the tag and date.
  source: https://github.com/orbital-materials/orb-models/releases/tag/v0.6.2
- version: v0.6.1
  date: '2026-03-11'
  breaking: null
  highlights: []
  note: Release notes not captured beyond the tag and date.
  source: https://github.com/orbital-materials/orb-models/releases/tag/v0.6.1
- version: v0.6.0
  date: '2026-03-04'
  breaking: false
  highlights:
  - GPU-accelerated graph construction via NVIDIA ALCHEMI Toolkit-Ops (up to 12x faster
    for large single systems; >100x for large batches of small systems).
  - TorchSim wrapper for batched optimisation and simulation.
  - Alchemi-based D3 dispersion correction module.
  source: https://github.com/orbital-materials/orb-models/releases/tag/v0.6.0
- version: v0.5.5
  date: '2025-08-21'
  breaking: false
  highlights:
  - Release of the OrbMol potentials, trained on Open Molecules 2025 (OMol25) — over 100M
    high-accuracy DFT calculations; explicit total charge and spin multiplicity can now be
    passed as input.
  source: https://github.com/orbital-materials/orb-models/releases/tag/v0.5.5
- version: 0.5.4
  date: '2025-04-30'
  breaking: null
  highlights: []
  note: Part of the April 2025 Orb-v3 release series (arXiv:2504.06231).
  source: https://github.com/orbital-materials/orb-models/releases/tag/0.5.4