NAISS and WASP in joint project to train quality Swedish LLM

_{20 February 2026}

Today the government presented its AI strategy. Among the highlights was the announcement of a new project initiated by the Knut and Alice Wallenberg Foundation (KAW) that will develop high-quality Swedish large language models (LLM) by training on the Berzelius supercomputer in Linköping, in a collaboration involving WASP – Wallenberg AI, Autonomous Systems and Software Program and NAISS teams.

While NAISS and the Mimer AI Factory are also committed to invest in fully open models trained on freely available data, a unique aspect of this project is that the rights holders (publishing houses, news media, authors) are directly involved. They will retain copyright to their data, but the project will enable research on how good models can become with access to copyrighted data. The government will similarly support the project by supplying large quantities of public data, something that Sweden has in abundance.

“This project is fantastic example of the impact WASP is having on AI in Sweden. It would not have been possible without a strong commitment to fund the Berzelius supercomputer. The support from the Swedish Research Council and the government for the Mimer AI Factory has also been vital”, says NAISS Director Erik Lindahl.

“It is of paramount importance to do this in close cooperation with copyright holders such that we build on the most valuable Swedish language resources available, advance research, but also ensure they are fairly compensated for commercial LLM applications building on their work.”

The LLM training will start immediately. According to KAW Executive Director Sara Mazur, the goal is to create a safe, ethical and sustainable language model that not only speaks and writes Swedish but also understands and incorporates the Swedish context in its output.

Researchers from several major Swedish universities, among them Linköping University, Chalmers University of Technology and KTH Royal School of Engineering, will also be part of the project.