Microsoft Research has released Skala 1.1, an upgraded deep-learning model for density functional theory (DFT) that achieves near-hybrid functional accuracy while maintaining the computational efficiency of simpler methods. The model, trained on 2.5 times more data than its predecessor, is now available in CP2K and being integrated into four additional major computational chemistry platforms: Psi4, FHI-aims, ORCA, and VASP. This expansion marks a significant shift toward making cutting-edge AI-powered molecular simulation accessible to the thousands of scientists and engineers who rely on these tools daily.

What Makes Skala 1.1 Different From Traditional Chemistry Models?

Density functional theory is the computational workhorse behind drug discovery, materials science, catalysis research, and energy technology development. For decades, chemists have relied on mathematical "functionals" to approximate how electrons behave in molecules. The problem: older functionals require researchers to choose between accuracy and speed. Skala takes a different approach by using machine learning to learn better approximations directly from high-accuracy quantum chemistry data.

Skala 1.1 achieves a weighted average error of 2.8 kilocalories per mole on GMTKN55, a widely used benchmark suite spanning 55 categories of chemistry including thermochemistry, reaction barriers, and noncovalent interactions. This level of accuracy surpasses today's leading global hybrid functionals while retaining the computational efficiency of semi-local methods, meaning researchers can run simulations faster without sacrificing precision.

Unlike the traditional "functional zoo," where new methods accumulate without replacing older ones, Skala follows a continuous-improvement philosophy. Each new release is designed to supersede the previous version, incorporating better training data, improved model architectures, and refined strategies. This approach allows the model to improve systematically with each generation while maintaining the same practical computational cost.

How Is Microsoft Expanding Access to Skala Across the Scientific Community?

The real-world impact of any advanced computational tool depends on accessibility. A breakthrough that only works in one software package reaches a fraction of its potential audience. Recognizing this, Microsoft Research has prioritized integrating Skala into the major platforms where computational chemists already work.

• CP2K Integration: Skala is now available in CP2K, a 25-year-old open-source package widely used for large-scale molecular dynamics and high-accuracy electronic-structure methods, expanding DFT accuracy while preserving computational efficiency at scale.
• Psi4 Integration: Microsoft is actively integrating Skala into Psi4, an essential platform for molecular electronic-structure research, which combined with the PySCF-based Skala Community Edition will make Skala available in three widely used open-source quantum chemistry packages.
• Commercial Software Partnerships: Microsoft is working with developers of FHI-aims, ORCA, and VASP to make Skala broadly accessible across the major commercial and open-source platforms used in computational chemistry and materials science.

Beyond software integration, Microsoft Research is introducing a "living benchmark" that will continuously track the computational performance of successive Skala releases across different software packages and hardware platforms. This transparent, regularly updated reference will help the scientific community measure progress and accelerate optimization efforts.

What Data Improvements Power Skala 1.1's Accuracy Gains?

The jump in accuracy from Skala 1.0 to 1.1 stems from a major expansion of the Microsoft Research Accurate Chemistry Collection (MSR-ACC), a large-scale repository of high-accuracy quantum chemistry reference data generated using expensive wavefunction methods. For Skala 1.1, researchers added new categories including electron affinities and noncovalent clusters, increasing both the size and diversity of the training dataset.

This data-driven approach demonstrates a fundamental principle in machine learning: better models emerge from better training data. By expanding the diversity of chemistry problems represented in the training set, Skala 1.1 learned to generalize more effectively across thermochemistry, reaction kinetics, and molecular structure prediction. The model also now provides highly accurate electron densities, dipole moments, and molecular geometries, not just energy predictions.

Steps to Evaluate and Deploy Skala in Your Research Workflow

• Community Edition First: Start with the open-source Skala Community Edition, built on GPU4PySCF and integrated with ASE (Atomic Simulation Environment), which enables researchers to evaluate and apply Skala with minimal effort while benefiting from highly optimized CPU and GPU performance.
• Check Your Software Platform: Verify whether your primary computational chemistry software (CP2K, Psi4, FHI-aims, ORCA, or VASP) has Skala integrated or is in the process of integration, as this determines the ease of adoption within your existing workflows.
• Monitor the Living Benchmark: Regularly consult Microsoft's continuously updated benchmark to understand how new Skala releases perform on your specific chemistry problems and computational hardware, helping you decide when to upgrade.
• Collaborate on Integration: If your preferred software platform hasn't yet integrated Skala, engage with the developer community or Microsoft Research to express interest, as community demand helps prioritize integration efforts.

The broader significance of Skala 1.1 lies in its demonstration that machine learning can systematically improve computational chemistry without requiring researchers to abandon the software tools they've built their workflows around. By meeting scientists where they already work, rather than asking them to adopt entirely new platforms, Skala has the potential to accelerate discovery across drug development, materials science, and catalysis research.

Microsoft Research's approach also signals a shift in how AI research is being deployed in scientific domains. Rather than publishing a paper and moving on, the team is investing in the infrastructure, partnerships, and benchmarking systems needed to ensure that improvements in AI actually reach the laboratories and companies that can benefit from them. This commitment to accessibility and continuous improvement positions Skala as a foundation for the next generation of computational chemistry workflows.