Beyond Big Tech: Switzerland’s New Open-Source AI Is Built for the Public Good

Switzerland has launched Apertus, an open-source large language model created as public infrastructure and jointly developed by EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS). The project is supported by Swisscom and the Hugging Face platform and is pitched as a public utility to sit alongside electricity and highways in terms of accessibility.

Apertus is fully open, with its training process, source code and datasets disclosed, and designed to comply with Swiss data protection and copyright rules, alongside Europe’s regulatory standards. Technically, the two model sizes are 8 billion and 70 billion parameters, trained on a dataset of 15 trillion tokens across more than 1,000 languages, with about 40% of the corpus non-English including Swiss German and Romansh.

All training data is drawn from public sources under machine-readable exclusion protocols to avoid issues surrounding data scraping. The article notes that the Swiss financial sector could benefit from an independent, domestically governed model aligned with strict banking secrecy and personal data protection. according to Swissinfo[.]ch