open force field

An open and collaborative approach to better force fields

Force fields

Force fields released by the Open Force Field Initiative can be accessed in the GitHub repository.

The files available for download are in SMIRKS Native Open Force Field (SMIRNOFF) format. Details about this new format are documented in our recent publication (or preprint). Examples of loading and using these force fields for a variety of use cases are available in our documentation.

If you experience any issues with the new force field, please be sure to email or raise an issue on the appropriate GitHub repository.



Sage (OpenFF 2.y.z) is the codename for the second generation of the small molecule force fields released by the Open Force Field Initiative. This line of force field release contains our first set of retrained Lennard-Jones parameters in addition to continued improvements to the valence parameters. All scripts, inputs and the results generated as part of the training the Sage line of OpenFF force fields can be found in the linked GitHub repositories.

openff 2.2.0

Release date: April 18, 2024

Updates to openff-2.1.0 which include improvements to small ring angles and sulfamide parameters. Contributions for this release include changes from Alexandra McIsaac and Pavan Behara, benchmarking results and infrastructure from Alexandra McIsaac, Brent Westbrook, Matthew Thompson, Jeffrey Wagner, Meghan Osato, Hannah Baumann, and Lily Wang, with valuable feedback from Lily Wang, Trevor Gokey, Chapin Cavender, David Mobley, Lee-Ping Wang, Christopher Bayly, Bill Swope, and the OpenFF team.

Find more details in the release notes, including how to re-run optimization and datasets used in training stored in sage-2.2.0 repo.

GitHub DOI

openff 2.1.0

Release date: May 2, 2023

Updates to openff-2.0.0 which include improved chemical perception for sulfonamides and phosphates, extended training set coverage which now contains training data from Gen2 as well as Gen1 datasets, and improved fitting procedures with the use of a physically intuitive starting point from modified seminario and including dihedral deviations in optimized geometry targets, and optimizing impropers as well. Contributions for this release include changes from Pavan Behara, Trevor Gokey, Chapin Cavender, Josh Horton, and valuable feedback from David Mobley, Hyesu Jang, Lily Wang, Christopher Bayly, Daniel Cole, and the OpenFF team.

Changes can be broadly classified as:

  • Chemical typing related
    • sulfonamides
    • phosphates
    • bridgehead nitrogens
    • bridgehead carbons
    • groups with delocalized charges
  • Fitting procedure related
    • use of physically intuitive bonds and angles from modified-seminario as a starting point
    • data-driven values for improper torsions
    • including dihedral deviations from optimized geometries to better resolve torsion parameters
    • broader coverage of parameters with extended training targets

Find more details in the release notes, including how to re-run optimization and datasets used in training stored in sage-2.1.0 repo.

GitHub DOI

openff 2.0.0

Release date: August 18, 2021

OpenFF 2.0.0 was created using a multi-stage optimization starting from the OpenFF 1.3.0 force field:

  • A select set of the vdW parameters were trained against a set of experimental mass density and enthalpy of mixing measurements sourced from the NIST ThermoML archive.
  • A select set of the bond length, bond force constant, equilibrium angle, angle force constant and torsion barrier height parameters were trained against a set of QC computed optimized geometries and torsion profiles, whereby the output force field of the first step was used as the starting point.

Find more details in the release notes, including how to re-run optimization and datasets used in training stored in openff-sage repo.

Instructions for running each optimization can be found in the individual optimization directories.

GitHub DOI



Parsley (OpenFF 1.y.z) is the codename for the first generation of optimized small molecule force fields from the Open Force Field Initiative. Parsley force fields represent a refit of the AMBER-lineage smirnoff99Frosst to large sets of quantum chemical data, and assign parameters using direct chemical perception, using the SMIRNOFF specification. Read more about the most important aspects of fitting and benchmarking procedures for Parsley in the blog post. See more information about naming and semantic versioning here.

All currently available versions are listed below. Each force field is currently available in two forms - with and without bond constraints to hydrogen. The default version of each force field is suitable for typical molecular dynamics simulations with constrained bonds to hydrogen. The unconstrained version of each force field should be used when single-point energies are a major concern (e.g. geometry optimizations) and when comparing the force field to QM data.

openff 1.3.1

Release date: June 21, 2021

This release is a bugfix for OpenFF 1.3.0 that corrects an issue with the tetrahedral geometry of certain sulphonamides (especially primary sulphonamides).

Full details are available in the GitHub release notes

GitHub DOI

openff 1.3.0

Release date: October 22, 2020

This minor release contains a fix of amide-related issues; (1) a poor performance of v1.2 in reproducing amide torsional energy profiles and (2) absence of appropriate torsion parameters for dialkyl amides. Full details are available in the GitHub release notes.

GitHub DOI

openff 1.2.1

Release date: September 9, 2020

This bugfix release manually changes two bond force constants (b24 and b27) to resolve an issue seen in propyne substituents when using hydrogen mass repartitioning with a 4 fs timestep. Full details are available at and in the reproduce_propyne_hmr.ipynb notebook in the release assets on GitHub.

GitHub DOI

openff 1.2.0

Release date: June 1, 2020

This Parsley update contains the following improvements:

  • Dramatically improved QM fitting dataset increased parameter quality across the board, for almost all molecules tested;
  • Improved treatment of phosphonate groups;
  • Substantial improvements to torsions for molecules with single bonds between a divalent nitrogen and a trivalent nitrogen.

Read more details about this force field update in the release notes.


Datasets used for training and initial benchmarking of OpenFF 1.2.0 force field are listed below:

DatasetSMILES  PDFStructures
FittingSMIPDFRelease tarball
Initial benchmarkSMIPDFRelease tarball
Lim and Mobley benchmark (in preparation, updates possible)SMIPDFSDF

The SMILES and PDF-format records are entirely derived from the structure files and may be updated without notice. If discrepancies are found, the structure files should be considered the authoritative reference.

GitHub DOI

OpenFF 1.1.1

Release date: April 15, 2020

This bugfix release contains the following changes: (1) Addition of monatomic ion LibraryCharges.

GitHub DOI

OpenFF 1.0.1

Release date: April 14, 2020

This bugfix release contains the following changes: (1) Addition of monatomic ion LibraryCharges.

GitHub DOI

OpenFF 1.1.0

Release date: March 3, 2020

The first Parsley update contains results of valence parameter re-fitting, with small modifications in the input force field. The main highlights include:

  • Addition of two new bond terms and one angle term;
  • Modification of periodicity for N-N rotation;
  • Addition of some proper and improper torsions.

Read more details about the modified parameters in the release notes or in the blog post.

GitHub DOI

OpenFF 1.0.0

Release date: October 12, 2019

The first official small molecule force field release from the Open Force Field Initiative. Fitting and benchmarking details are described here.

GitHub DOI



A minimalist AMBER-compatible general small molecule force field with excellent coverage of druglike chemical space, using the SMIRNOFF direct chemical perception specification.

This force field achieves comparable accuracy to GAFF in physical properties, but with a decrease in complexity from over 6000 lines of parameters to only ~300 lines using the SMARTS-based direct chemical perception possible with SMIRNOFF. Read more about the latter and creation of smirnoff99Frosst in the publication or in the preprint.

smirnoff99Frosst was used as a starting point to generate the first optimized Open Force Field (Parsley) while building automated infrastructure for rapid parameterization of accurate small molecule force fields.

GitHub DOI