NOVEMBER 2025
Architecting Digital Sovereignty
We are pleased to welcome you to this first newsletter from the Natural Language Understanding (NLU) team at AI Sweden to provide our community of researchers, developers, data managers, and applied AI professionals with technical insights into our ongoing work.
In this first edition, we share updates on our work with OpenEuroLLM and an interview with our Senior Data Lead, Danila Petrelli. On November 19, we host a seminar where she will talk more about data at AI Sweden's Stockholm office. Find links to this and other upcoming events at the end of the newsletter.
From National Models to European Flagship Projects
The NLU team has been actively engaged in LLM construction and research since 2021. Our initial high-profile work centered on the development of GPT-SW3, where the team was at the forefront of this large-scale model's creation.
Since then, we've significantly expanded both our internal capacity and our collaborations. Our model hub has expanded to over 40 models, containing both foundational base models and task-specific fine-tuned models.
Crucially, we've moved beyond focusing on isolated national projects to collaborating with fellow institutes and universities across Europe to build the next generation of large-scale language models within research projects such as TrustLLM and EuroLingua-GPT.
Our ambition for the coming years is to disseminate knowledge and insights from the team's contributions to OpenEuroLLM, the EU’s flagship project, as well as other key initiatives we are participating in.
AI Sweden's overarching goal is to accelerate the application of AI within Sweden for the benefit of our society, our competitiveness, and for everyone living in Sweden. Our team provides critical expertise in LLM architecture, high-performance computing (HPC) environments, development of large-scale LLMs, and critical development of data governance. Therefore, a priority is to establish transparency and advance the general understanding of this technology. By making model weights and training methodologies accessible, we aim to fortify Sweden’s and Europe's digital sovereignty and contribute to the democratization of language models on a broader scale.
