Skala 1.1 deep-learning DFT functional gets sharper and spreads to major chemistry codes
Microsoft Research's deep-learning exchange-correlation functional just got measurably better: trained on 2.5x more data, Skala 1.1 sharpens accuracy on thermochemistry, reaction kinetics, and structure prediction. The bigger move is reach — it now runs in CP2K and is being wired into Psi4, FHI-aims, ORCA, and VASP, the codes chemists actually use. A living benchmark now tracks how each successive release performs, so the field can measure the march toward predictive DFT rather than take it on faith.