Microsoft's Skala 1.1 tops 32 of 55 GMTKN55 categories, integrating into CP2K, ORCA, VASP
Microsoft Research · rss · 2026-08-21
Microsoft Research released Skala 1.1, its deep-learning exchange-correlation functional for DFT. Trained on 2.5x more data than the first public version, it achieves a weighted average error of 2.8 kcal/mol on the GMTKN55 benchmark and ranks first in 32 of 55 categories—beating the most expensive global hybrid functionals at the computational cost of a meta-GGA, while also delivering accurate electron densities, dipole moments, and molecular geometries.
On accessibility, Skala is now available in open-source CP2K and is being integrated into Psi4, FHI-aims, ORCA, and VASP. The MSR-ACC reference dataset gained new categories such as electron affinities and noncovalent clusters, and Microsoft introduced a living benchmark tracking the computational performance of successive Skala releases across packages and hardware.
Skala follows a continuous-improvement philosophy: each release supersedes the previous one, gaining accuracy as data and training scale while keeping computational cost constant—unlike the traditional "functional zoo" of accumulating functionals.
More from Research
- Transformer trained on flying bat recordings yields hippocampus 'digital twin' revealing hidden structure — patrickmineault · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- MIT PhD Sharut Gupta joins DeepMind; Isola hails her representational alignment defense talk — phillip_isola · 2026-10-16
- 36 primitives, just 10 parameters each: a remarkably precise T-pose body fit — letandrewcook · 2026-10-16
- Interp researcher pushes back: if you can't simulate a neuron, you haven't explained it — nabla_theta · 2026-10-16
- MLP-Splatting (ECCV 2026) code released: MLPs as light-field primitives instead of Gaussians — AjdDavison · 2026-10-16