Skip to main content

GPT-SW3

GPT-SW3 is the first large-scale generative AI model for the Nordic languages. It was developed in a collaboration between AI Sweden, RISE, and WASP WARA Media & Language.

Scrabble tiles tumbling in the air, each spelling out the letters and symbols of GPT-SW3

The launch of GPT-SW3 interpreted by OpenAI's DALL-E 3

GPT-SW3 is available in six different sizes: 126 million, 356 million, 1.3 billion, 6.7 billion, 20 billion, and 40 billion parameters. The model received significant international attention and gave AI Sweden a position in Europe and globally, which in turn has led to AI Sweden currently participating in several European projects. The largest of these is OpenEuroLLM, where around twenty stakeholders are collaborating to develop a family of large language models for the European languages.

GPT-SW3 was launched in two stages. First, a pre-release was made available for research purposes, and then in November 2023, it was released under an open license that permitted broad use in both the public and private sectors. Today, the model is no longer available for open download. It is possible to apply for access for research purposes according to the process below.

Access to GPT-SW3

If you are a researcher at a European university or research institute, you can apply for access to the model in the following way:

  • Search for the model you wish to use on HuggingFace.
  • Read through the terms of the license.
  • Fill in the requested information.
  • Consent to sharing your information with us and accept the terms of the license.
  • By consenting, you share your username, name, email address, and affiliated organization with AI Sweden.
  • The process is manual, and you will receive an email response once your application has been reviewed and evaluated.

The application requires you to have an account on HuggingFace and to be working at a European university or research institute. Applications are submitted once per model and are processed on an individual basis.

Ongoing language projects at AI Sweden

OPEN EURO LLM project co-funded by the European Union

OpenEuroLLM

Europe's leading AI companies and research institutions combine their forces and expertise to develop next-generation open-source language models in an unprecedented collaboration to advance European...
People working on laptops in a collaborative office environment.

A shared digital assistant for the public sector

The shared digital assistant for the public sector is a collaboration between Swedish authorities, municipalities, regions, and industry, coordinated by AI Sweden. The objective is to promote national...

Read more

Woman in grey blazer next to Open Euro LLM logo on a blue background.
It has now been a year since the launch of OpenEuroLLM, one of Europe's most ambitious AI...
Rows of black server racks with colorful cables in a bright, modern data center with glass floor panels. Eu-flag with the text 'Co-funded by the European Union'
OpenEuroLLM is set to develop high-performance language models for Europe. The major European...
Nina Ökvist and Magnus Sahlgren (composite image)
The availability of large, open language models, along with the expertise required to develop...
Danila Petrelli
The accelerating use of large language models is unlocking enormous value for organizations. But to...

For more information

Subscribe to the NLU newsletter, or contact

Jonas Lind portrait photo
Jonas Lind
Head of research, NLU, PhD