Switzerland launches Apertus, an open multilingual AI model

Switzerland has launched its first large-scale open-source language model, Apertus, developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre. Trained on the powerful Alps supercomputer in Lugano, Apertus is designed to set a new standard for transparency and multilingual inclusivity in Europe’s AI landscape.

The model comes in two sizes, 8 billion and 70 billion parameters, and supports over 1,000 languages, with 40% of its training data drawn from non-English sources. That allows it to handle underrepresented languages such as Swiss German and Romansh more effectively.

Unlike proprietary AI systems, Apertus is fully open. Its architecture, training data recipes, model weights, and documentation are publicly accessible.

The model can be downloaded from Hugging Face or accessed via Swisscom’s sovereign Swiss AI platform, with both research and commercial use permitted under a permissive license. Developers highlight that its design ensures compliance with Swiss and the EU regulations, with careful filtering of training data for quality and ethical standards.

The release comes just ahead of the Swiss {ai} Weeks hackathons, where developers and researchers will put Apertus to the test. Organisers describe the model as more than a research breakthrough, framing it as a tool to drive innovation across society and industry.

Looking ahead, the Apertus project aims to expand its family of models, improve efficiency, and develop specialised versions for fields like law, health, climate, and education, further strengthening Switzerland’s role in shaping open, public-benefit AI.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!