Broadening access to Skala creates a faster path to predictive DFT

Details

Microsoft Research has updated Skala, its deep-learning exchange-correlation functional for density functional theory (DFT), to version 1.1. Trained on 2.5 times more data than its predecessor, Skala 1.1 ranks first in 32 of 55 GMTKN55 categories — the standard benchmark for molecular simulation accuracy — covering thermochemistry, reaction kinetics, and molecular structure prediction. It delivers these results at the computational cost of a meta-GGA functional, a fraction of what expensive global hybrid functionals require.\n\nDFT is the computational backbone of materials science, drug discovery, catalysis, and energy research. The model is now available in CP2K and is actively being integrated into Psi4, FHI-aims, ORCA, and VASP — the software packages where most computational chemists actually work. This moves Skala from a research artifact to a tool accessible within existing workflows.\n\nMicrosoft is also introducing a living benchmark that will track successive Skala releases' computational performance over time, providing transparent community evaluation of future improvements. This continuous-improvement model distinguishes Skala from one-off research publications.\n\nFor AI practitioners, this represents one of the clearest examples in 2026 of AI models delivering measurable, independently benchmarkable improvements over traditional physics-based methods in a high-stakes scientific domain.