EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) have released Apertus, a large-scale, open, multilingual large language model. The project name Apertus is Latin for open and the team says the entire development process is openly accessible. That includes model architecture, model weights, intermediate checkpoints, training data and training recipes, all accompanied by comprehensive documentation and source code.
The release is positioned as a tool for research, industry and public examination rather than a conventional technology transfer to a single product. Thomas Schulthess, director of CSCS and professor at ETH Zurich, frames the project as a driver of innovation and a way to strengthen expertise across research, society and industry. Imanol Schlag, technical lead of the LLM project and research scientist at ETH Zurich, says Apertus is built for the public good and designed around multilingualism, transparency and compliance.
Access will be provided through several routes. Swisscom will offer a dedicated interface for participants at upcoming Swiss Artificial Intelligence Weeks hackathons and make the model available to Swisscom business customers via its sovereign Swiss artificial intelligence platform. For users outside Switzerland, access will be offered through the Public Artificial Intelligence Inference Utility. The research team released the model under a permissive open-source license that also allows commercial use and says future work will expand the model family, improve efficiency, explore domain-specific adaptations for law, climate, health and education, and add capabilities while maintaining reproducibility and transparency.
