AI Sweden with its initiative "A shared digital assistant for the public sector" is among just a handful of organizations worldwide granted early access to test and evaluate OpenAI’s new open models ahead of public release.
![]()
Through early feedback from organizations like AI Sweden, we're together able to provide powerful, flexible tools that make it easier than ever to build, innovate, and scale.
![]()
Dmitry Pimenov
Product lead at OpenAI
AI Sweden's project, A shared digital assistant for the public sector, also known as the Svea project, exemplifies the kind of groundbreaking, socially impactful AI initiative that OpenAI aims to support by opening access to its new open models. The collaboration also means that AI Sweden will help assess how these open models perform in real-world, high-impact settings.
"Our open models let developers—from solo builders to public sector organizations—run and customize AI on their own infrastructure, unlocking new possibilities and use cases," said Dmitry Pimenov, product lead at OpenAI. "Through early feedback from organizations like AI Sweden, we're together able to provide powerful, flexible tools that make it easier than ever to build, innovate, and scale."
![]()
We welcome OpenAI's initiative to release these new open models. It marks an important shift that aligns perfectly with our approach to developing powerful open models that can be tailored to specific needs. The fact that AI Sweden has been given this opportunity is a direct result of the capabilities we've built over several years, through the development of Swedish and European large language models and the groundbreaking Svea project. This collaboration will strengthen Swedish innovation and digital sovereignty, and it offers us a unique opportunity to adapt and implement the very latest technology for Swedish needs.
![]()
Magnus Sahlgren
Head of Research, Natural Language Understanding, at AI Sweden
The Svea project unites 55 public sector organizations – including over 20 percent of Sweden's municipalities – in a unique effort to develop a prototype for a shared digital assistant for the public sector.
![]()
By combining open source technology with secure infrastructure from Swedish infrastructure provider Airon, the Svea project demonstrates how the public sector can build powerful and cost-effective AI solutions while retaining full control over its own data.
![]()
Jonatan Permert
AI Transformation Strategist at AI Sweden and Project Manager for the Svea project
What makes the project particularly innovative is its collaborative data effort: over 300 employees from across participating organizations have contributed thousands of working hours to create a database with over 200,000 annotated data points. The capability and expertise AI Sweden has developed is a key reason to why AI Sweden is now in the position to maximize the utility of OpenAI’s powerful new open models.
"By fine-tuning OpenAI's open models, we can enable them to better understand Swedish administrative terminology, legal language, municipal workflows, and document structures within the public sector. This makes it possible to build powerful, sovereign, and cost-effective AI solutions tailored to the needs of Swedish organizations," said Magnus Sahlgren.
AI Sweden's project, "A shared digital assistant for the public sector," is a collaborative initiative involving Swedish authorities, municipalities, regions, and industry players. The goal is to create conditions for common AI solutions in the public sector.
The project, which takes place in two stages between 2024 and 2026, involves over 50 municipalities, regions, and government agencies that co-finance the project alongside Vinnova. Preparations are currently underway for a third stage in 2026.
Within the project, Svea is being developed as a prototype of a digital assistant in the form of a web-based chat interface. Built on open source and Swedish infrastructure, the prototype is designed to be model-agnostic, working with all open large language models. Svea is being developed to, among other things, summarize texts, improve written materials, classify documents, create decision-making bases in case management processes, and provide personalized citizen support.
Continue exploring