Switzerland just dropped something big in the AI world. A group of institutions—including EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS)—has launched a new open model called Apertus (Latin for “open”). And they really mean it: every detail about its design and training is open to the public.
So what can you do with it? Think chatbots, translation apps, or education tools. Apertus comes in two versions—one with 8 billion parameters, another beefier one with 70 billion. Both can be grabbed directly from Hugging Face or through Swisscom, which is teaming up as a strategic partner. Since it’s open-source under a very flexible license, you can use it for research, teaching, or even commercial projects without much red tape.
Here’s the key difference: while many AI models only share surface-level details, Apertus lays it all out. Its architecture, training data, and full documentation are up for inspection. Martin Jaggi from EPFL, part of the steering committee, put it simply: this is a blueprint for building AI that’s trustworthy, sovereign, and inclusive. Regular updates are on the way from engineers and researchers at EPFL, ETH Zurich, and CSCS.
Thomas Schulthess from CSCS sees Apertus as more than just a research-to-product pipeline. Instead, he calls it infrastructure—something designed for long-term impact across science, society, and industry.
The model’s training is also pretty unique. It chewed through 15 trillion tokens across more than 1,000 languages, with around 40% of the data coming from non-English sources. That includes languages that usually get overlooked in AI, like Swiss German and Romansh.
As Imanol Schlag, ETH Zurich research scientist and technical lead, explains: Apertus isn’t just another open model. It’s one of the few at this scale that builds multilingualism, transparency, and compliance into its very foundation.
Swisscom is already rolling it out on its sovereign AI platform, which Research Director Daniel Dobos says strengthens Switzerland’s digital independence while promoting responsible AI.
For developers itching to try it out, there are options. If you’ve got the right servers or cloud resources, you can download Apertus today. If not, Swiss {ai} Weeks (running until October 5, 2025) offers hands-on access, especially for hackathon participants. Swisscom business clients can tap into it immediately, and international users will soon have access through the Public AI Inference Utility.
Joshua Tan, who leads that utility, calls Apertus the best proof yet that AI can be public infrastructure—like roads, water, or electricity—built by and for the public.
Transparency is at the heart of this release. Training data, weights, and even intermediate checkpoints are available. Everything followed Swiss data protection rules, copyright law, and the EU AI Act’s transparency requirements. Sensitive personal data was filtered out, websites’ opt-outs respected, and ethical safeguards applied.
Looking ahead, this isn’t the end of Apertus but the start of something bigger. Antoine Bosselut of EPFL describes it as the beginning of a long-term effort to keep AI powerful, open, and trustworthy. Plans include expanding the model family, making it more efficient, and tailoring versions for fields like law, health, climate, and education—while keeping transparency front and center.
