Switzerland launches “Apertus,” an open-source AI model trained on public data
Countries are racing to build “sovereign AI,” and Switzerland is the latest country to release its own open-source AI model for anyone to use.
True to its tradition of neutrality and independence, Switzerland is moving closer to securing its own “sovereign AI.”
Wary of reliance on American and Chinese startups for access to cutting-edge AI, countries are taking steps to develop domestic AI infrastructure.
That includes the AI models that run on the data centers inside a country’s borders. A group of Swiss universities teamed up to develop “Apertus” (Latin for “open”), a large multilingual language model trained excessively on public data.
The model was released by the Swiss National Supercomputing Centre, along with universities École Spéciale de Lausanne and ETH Zurich.
The Apertus 2 model is roughly comparable to Meta’s last-gen Llama 3 AI model, and was trained on 15 trillion tokens and 1,000 languages including Swiss German and Romansh, which is a national language of Switzerland.
Switzerland telecom company Swisscom AG has partnered with AI chip leader Nvidia to build out its domestic AI infrastructure for Swiss businesses.
The Swiss National Supercomputing Centre is home to “Alps,” an AI computing cluster filled with over 10,000 Nvidia H200 GPUs that was built for Swiss researchers.
Having a free model that was built domestically, with full transparency, will help domestic businesses that use the model to comply with the strict data protection and intellectual property laws in Europe.
