Switzerland Releases its First Large-Scale, Open, Multilingual Language Model

Apertus, a fully open, transparent, multilingual language model, developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS), was released this week.

Apertus, Latin for “open”, is a fully transparent AI model with openly accessible architecture, weights, training data, and documentation. Available via Swisscom or Hugging Face, it comes in 8B and 70B parameter sizes, released under a permissive open-source license for education, research, commercial, and societal use.

Apertus is a fully open language model, enabling researchers, professionals, and enthusiasts to adapt it to their needs and inspect every part of the training process—unlike models that reveal only select components.

With its open approach, EPFL, ETH Zurich, and CSCS aim to drive innovation and strengthen AI expertise across research, industry, and society. Trained on 15 trillion tokens in over 1,000 languages, including many underrepresented ones like Swiss German and Romansh, Apertus provides both foundational technology and infrastructure to foster broad economic impact.

Setting up Apertus is straightforward for experienced users, though additional infrastructure like servers or cloud platforms is needed for practical use. The upcoming Swiss {ai} Weeks will offer developers hands-on experience to test the model and provide feedback. Swisscom will provide a dedicated interface for hackathon participants, and its business customers can access Apertus via Swisscom’s sovereign Swiss AI platform. Globally, the Public AI Inference Utility makes Apertus accessible as part of a public AI movement, showcasing how AI can serve as public infrastructure, says Joshua Tan, Lead Maintainer.

Apertus is built for full transparency, ensuring reproducibility of the training process. The research team has released model weights, intermediate checkpoints, source code, datasets, and comprehensive documentation under a permissive open-source license that allows commercial use. Developed in line with Swiss data protection and copyright laws, as well as EU AI Act transparency requirements, Apertus uses only publicly available data, respects machine-readable opt-outs, and removes personal or undesired content to maintain ethical standards.

Future Apertus versions will expand the model family, enhance efficiency, explore domain-specific adaptations in areas like law, climate, health, and education, and add new capabilities while maintaining transparency standards.