"A group of Swiss institutions has released Apertus, a new large language model (LLM) designed to advance research, education, and commercial applications. Built by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS), Apertus is entirely open, with architecture, training data, and documentation publicly available. The model comes in two versions - an 8-billion parameter and a 70-billion parameter version - both licensed for unrestricted use in research, education, and commercial projects."
"Apertus was trained on 15 trillion tokens across 1,000+ languages, with 40% of its dataset in non-English content. It includes underrepresented languages such as Swiss German and Romansh, making it one of the most inclusive LLMs to date. " Apertus is built for the public good," said Imanol Schlag, technical lead and Research Scientist at ETH Zurich. " It is the first of its kind to embody multilingualism, transparency, and compliance as foundational design principles. ""
"" With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed," said Martin Jaggi, Professor of Machine Learning at EPFL. Regular updates will be managed by engineers and researchers from CSCS, ETH Zurich, and EPFL. Thomas Schulthess, Director of CSCS, added that Apertus is " a driver of innovation and a means of strengthening AI expertise in research, society and industry." He emphasized that the project aims to create long-term infrastructure, rather than a one-time technology transfer."
Apertus is an open large language model developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS). The model is released in 8-billion and 70-billion parameter variants with architecture, training data, and documentation publicly available. Licensing permits unrestricted use in research, education, and commercial projects. Access is offered via Hugging Face and through Swisscom's sovereign AI platform, with testing during Swiss {ai} Weeks through October 5, 2025, and business availability for Swisscom customers. Apertus was trained on 15 trillion tokens spanning over 1,000 languages, with 40% non-English data and inclusion of Swiss German and Romansh. Ongoing updates will be managed by CSCS, ETH Zurich, and EPFL engineers and researchers.
Read at Medium
Unable to calculate read time
Collection
[
|
...
]