Switzerland has joined the large language model race by introducing its own national artificial intelligence, Apertus. Developed by scientists from state institutions, this model is public infrastructure and fully accessible to everyone, as its source code is open.

The Apertus model was trained on a massive dataset of 15 trillion tokens in over 1,000 languages. Notably, 40% of the data is non-English, which makes it particularly useful for researchers and companies working with various languages. Apertus is currently available in two versions—with 8 and 70 billion parameters—via the Swisscom and Hugging Face platforms.

ORIENT