Microsoft Opens Skala 1.1 to More Chemistry Codes
Microsoft Research says Skala 1.1, its updated deep-learning exchange-correlation functional, improves accuracy in density functional theory while moving into more of the software used by computational chemists. The announcement combines a model update, native software integrations and a living benchmark intended to track performance over future releases. For researchers, the practical question is whether higher-accuracy DFT can become easier to run inside established scientific workflows rather than remaining tied to one release channel.Why Skala 1.1 Matters for DFT
Skala 1.1 is Microsoft Research’s latest step toward making density functional theory more predictive for molecular simulation. The group describes Skala as a deep-learning exchange-correlation functional, a component at the core of how DFT approximates the electronic structure of molecules and materials.The announcement frames DFT as a computational engine used across chemistry, materials science, catalysis, energy technologies and drug discovery. That matters because a more accurate functional has limited impact if it is difficult to run in the codes that laboratories and industrial teams already use. Microsoft Research’s pitch for Skala 1.1 is therefore twofold: better measured accuracy and broader availability across the computational chemistry ecosystem.
Accuracy Claims and Training Data
Microsoft Research says Skala 1.1 was trained on 2.5x more data than the first public version of Skala. The added training data comes from expansions of the Microsoft Research Accurate Chemistry Collection, a set of high-accuracy quantum-chemistry reference data generated with expensive wavefunction methods.The source says new categories include electron affinities and noncovalent clusters, increasing both the size and diversity of the training set. On GMTKN55, a benchmark suite with 55 chemistry categories including thermochemistry, reaction barriers and noncovalent interactions, Microsoft Research reports a weighted average error of 2.8 kcal/mol for Skala 1.1.
The company also says the updated model improves performance on main-group thermochemistry, reaction kinetics and molecular structure prediction. It further claims Skala 1.1 surpasses leading global range-separated hybrid functionals while retaining the efficiency of a semi-local functional, and provides highly accurate electron densities, dipole moments and molecular geometries. Those are vendor-reported technical results, but they define the performance target Microsoft wants the wider DFT community to test.
Software Access Moves Beyond One Package
The accessibility news is that Skala is now available in CP2K and is being integrated into Psi4, FHI-aims, ORCA and VASP. Microsoft Research says this follows an earlier open-source community release built on GPU4PySCF and integrated with ASE.That expansion is significant because computational chemistry does not operate through a single dominant codebase. Different electronic-structure packages have developed around different numerical methods, hardware choices, research communities and industrial use cases. A functional that works only in one environment can be difficult to compare, validate or adopt at scale.
By targeting CP2K, Psi4, FHI-aims, ORCA and VASP, Microsoft Research is trying to place Skala closer to workflows already used for molecular electronic-structure research, large-scale simulation and materials science. The source describes the integrations as collaborative work with software developers rather than a standalone product rollout.
CP2K Integration and Validation
CP2K is the first named package in the announcement where Skala has been successfully integrated. Microsoft Research says the work was done in collaboration with the team of Prof. Thomas D. Kühne at the Center for Advanced Systems Understanding, known as CASUS.The source describes CP2K as an open-source package with more than 25 years of development, especially relevant for DFT simulations of large-scale systems and long-timescale molecular dynamics. Microsoft Research says Skala expands what is possible inside CP2K by improving DFT accuracy while preserving the computational efficiency needed for simulations at scale.
Validation is a major part of the CP2K case. Microsoft Research says it developed a suite of integration tests with the CP2K team to verify numerically correct and reliable results across computational settings. The announcement also points to a joint paper with the CASUS team on the molecular implementation of the machine-learned Skala exchange-correlation functional in CP2K through GauXC.
A Living Performance Report
Microsoft Research is also publishing a benchmarking harness and a living performance report for Skala. The purpose is to track computational performance across successive Skala releases, software packages and hardware platforms as implementations are optimized.The source says Skala can currently deliver performance comparable to semi-local meta-GGAs on CPUs, with an overhead that disappears for molecules with more than 20-30 atoms, and on GPUs. It also says performance is expected to change as new Skala releases, GauXC improvements and hardware-specific optimizations arrive.
A living benchmark is useful because accuracy alone is not enough for routine molecular simulation. Package developers need a common way to compare implementations, identify bottlenecks and verify that performance improvements do not break numerical behavior. For users, the benchmark can clarify whether a Skala implementation is mature enough for a given workload.
Conclusion
Skala 1.1 is not presented as a one-time breakthrough but as part of a continuously improving DFT approach. Microsoft Research says each Skala release is intended to supersede the previous one as more data, model architectures and training strategies become available while maintaining practical computational cost.The near-term test is adoption and reproducibility across established codes. If CP2K, Psi4, FHI-aims, ORCA and VASP integrations produce consistent results and usable performance, Skala could become easier to evaluate in real research settings. If not, the living benchmark should make the gaps visible. Either way, the announcement shifts Skala from a model-centered update toward a broader infrastructure effort for predictive DFT.
Sources
Editorial Team - CoinBotLab
