What is Apertus, Switzerland’s open AI model?
Apertus is Switzerland's first large-scale, fully open language model, released on 2 September 2025 by EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS). Everything about it - architecture, model weights, training data and methods - is publicly documented and free to use, including commercially.
What "fully open" means here
Many "open" models release only their weights. Apertus publishes the entire development process: source code, training data documentation, model weights including intermediate checkpoints, all under a permissive open-source licence. Anyone can inspect how it was trained and build on it.
Two sizes are available: 8 billion and 70 billion parameters. The smaller one suits individual use; both can be downloaded from Hugging Face or accessed through Swisscom's sovereign Swiss AI platform and the Public AI Inference Utility.
Why it was built
The model was trained on 15 trillion tokens across more than 1,000 languages, with 40% of the data non-English. That includes languages most LLMs handle poorly, such as Swiss German and Romansh.
The project is a deliberate counter-model to closed systems: "a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed," in the words of EPFL professor Martin Jaggi. It was developed with Swiss data protection law, Swiss copyright law and the EU AI Act's transparency obligations in mind - the training corpus uses only publicly available data, honours machine-readable opt-out requests (even retroactively), and is filtered to remove personal data before training.
Why it matters for Swiss organisations
Data control: because the weights are open, a company can run Apertus on its own infrastructure - in Switzerland - instead of sending data to a foreign API.
Compliance: full documentation of training data and methods makes audits possible in a way closed models do not allow.
Language coverage: native handling of Switzerland's four national languages, including Swiss German dialects.
It is a foundation, not a finished product: using it well still requires servers, integration and interfaces. But for organisations that need sovereign, inspectable AI, Switzerland now has a public baseline to build on.
Sources
ETH Zurich, "Apertus: a fully open, transparent, multilingual language model" (2 September 2025)
CSCS, "Apertus: a fully open, transparent, multilingual language model"
EPFL, "Apertus: a fully open, transparent, multilingual language model"
Last updated: 20 October 2026.
Wondering whether an open Swiss model like Apertus fits your organisation? We help you test it on your own data and infrastructure.
