Microsoft Research releases Skala 1.1, a deep-learning exchange-correlation functional with improved accuracy and expanded ecosystem support

Microsoft Research released Skala 1.1, an updated deep-learning exchange-correlation functional that improves accuracy and expands accessibility across the computational chemistry ecosystem. The update trains on 2.5× more data than its predecessor, delivering substantially higher accuracy across key molecular simulation challenges such as thermochemistry, reaction kinetics, and molecular structure prediction.

Skala 1.1 achieves a weighted average error of 2.8 kcal/mol on the GMTKN55 benchmark, which spans 55 categories of chemical problems including thermochemistry, reaction barriers, and noncovalent interactions. This level of accuracy surpasses leading global hybrid functionals while retaining the computational efficiency of a semi-local functional.

The model provides highly accurate electron densities, dipole moments, and molecular geometries, enabled by expansions to the Microsoft Research Accurate Chemistry Collection (MSR-ACC). New categories added to the training data include electron affinities and noncovalent clusters, increasing both the size and diversity of the dataset.

To broaden adoption, Skala 1.1 is now available in CP2K and is being integrated into Psi4, FHI-aims, ORCA, and VASP. These integrations aim to bring next-generation DFT accuracy to communities that rely on these widely used electronic-structure software packages.

Microsoft Research also introduced a living benchmark to track the computational performance of successive Skala releases. The benchmark provides a transparent and continuously updated reference for implementations across software packages and hardware platforms, helping the community measure and accelerate progress toward greater accuracy and efficiency.