Microsoft's Skala 1.1, a deep-learning density functional, brings an upgrade to the workhorse codes of computational chemistry

Microsoft Research released Skala 1.1, a deep-learning upgrade to the exchange-correlation math used in density functional theory (DFT), the workhorse method computational chemists use to simulate molecules and materials for drug discovery, catalysis, and battery design. The update ships with 2.5x more training data than the first public Skala release and, by Microsoft's own reporting, lifts performance on main-group thermochemistry, reaction kinetics, and molecular structure prediction.

The structural change is access, not novelty: Skala now runs inside CP2K, one of the codes working chemists actually use, and Microsoft has integrations in progress for Psi4, FHI-aims, ORCA, and VASP. As a learned exchange-correlation functional, it drops into existing pipelines rather than arriving as a separate AI tool.

Microsoft is also launching a "living benchmark" to track successive Skala releases. It is empty so far: no independent data has been added, and Microsoft's announcement is the only source for the accuracy claims. The accompanying preprint describes the method but does not resolve whether the 2.5x-data-driven gain transfers outside Skala's training chemistry.