MateriApps
A Portal Site of Materials Science Simulation

now 335 Apps

Inquiry / Application Request
Detailed search

Matbench Discovery

  • Openness:3 ★★★
  • Document quality:2 ★★☆

A benchmark framework for evaluating general-purpose, i.e., universal, machine learning potentials, along with a leaderboard based on those evaluations. Rankings are determined by a comprehensive assessment that considers the accuracy of predicted formation energy of materials, structural relaxation, and thermal conductivity. Recently, in addition to public research institutions such as universities, major companies like Meta, Microsoft, and Google have also joined the development of universal potentials, taking top positions on the leaderboard.

Information
Last Update:2025/11/10

Official site

https://matbench-discovery.materialsproject.org/

License

MIT License

Availability

Can be accessed by any modern web browser
Core Developers

See Github contributors page https://github.com/janosh/matbench-discovery/graphs/contributors
Target substance/model

Potentials on the leaderboard can be used for any material, but the benchmark mostly considers inorganic materials.
Related App
Document

Janosh Riebesell, Rhys E. A. Goodall, Philipp Benner, Yuan Chiang, Bowen Deng, Mark Asta, Gerbrand Ceder, Alpha A. Lee, Anubhav Jain, Kristin A. Persson. “Matbench Discovery — A Framework to Evaluate Machine Learning Crystal Stability Predictions.” arXiv, August 28, 2023. https://doi.org/10.1038/s42256-025-01055-1.
TOP