| Switzerland Releases Its Own Large Language Model |
| Written by Nikos Vaggalis | |||
| Tuesday, 30 September 2025 | |||
|
Code-named Apertus, it's open-source and multilingual trained on trillions of tokens. What makes it different? While other corporate models like OpenAI's are baptized open-source, that term refers only to their weights being released to public. Apertus is open source all the way, from the beginning to the end. Together with the weights, the developers have also released the source code, the training data and the development process. Apertus was trained on 15 trillion tokens across more than 1,000 languages including languages that have been underrepresented in LLMs, such as Swiss German, Romansh, and others. Power wise it is comparable to Meta’s Llama 3 model from 2024. The other unique characteristic of the model is that it respects privacy, in that it has been made to filter out personal information, trained only on publicly available data that adheres to Swiss and EU copyright laws. This last point will protect the Swiss makers from potential lawsuits and fines like those imposed on Meta and OpenAI due to them training their models on copyrighted works such as books. The openness of the model follows the recommendations of The Open Data Institute for a competitive Europe and specifically recommendation no2, Open Data As A Foundation, as we covered in The European Data And AI Policy Manifesto For A Stronger Europe: The future is Open. Open data, open source. Quote: The best possible foundation is open data, supported and sustained as data infrastructure. Only with this foundation will people, businesses and governments be able to realize the potential of data infrastructure across society and the economy. Along openness, ODI suggest to also foster data altruism; organizations that voluntarily share data for the benefit of society. Apertus is available in two sizes; 8 billion and 70 billion parameters. Both models are released under a permissive open-source license, allowing use in education and research as well as broad societal and commercial applications. They're available for downloading from Hugging Face.
More InformationRelated ArticlesThe European Data And AI Policy Manifesto For A Stronger Europe
To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin. 
Comments
or email your comment to: comments@i-programmer.info |
