In late summer 2025, a new large language model will be made publicly available by a consortium of Swiss research institutions. According to an announcement from ETH Zurich by authors Florian Meyer and Mélissa Anchisi, the model was co-developed by researchers at EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS). It aims to provide an open and transparent alternative to commercial AI systems.
The model will be fully open, meaning its source code, trained parameters, and training data will be publicly accessible. This transparency is intended to build trust, support scientific research, and enable regulatory compliance. “Fully open models enable high-trust applications and are necessary for advancing research about the risks and opportunities of AI,” said Imanol Schlag, a research scientist at the ETH AI Center.
A key feature of the new model is its extensive multilingual capability. It was designed to be fluent in over 1,000 languages. Antoine Bosselut of the EPFL AI Center explained that the team emphasized making the model “massively multilingual from the start”. The training data included text from over 1,500 languages, with approximately 40 percent of the content being non-English.
To meet different user needs, the model will be released in two sizes: an 8-billion-parameter version and a 70-billion-parameter version. The number of parameters in a model reflects its capacity to process information and generate complex text. The developers state that the larger version will be among the most powerful fully open models available globally.
The training was conducted on the “Alps” supercomputer at CSCS in Lugano, which uses 100 percent carbon-neutral electricity. “Training this model is only possible because of our strategic investment in ‘Alps’, a supercomputer purpose-built for AI,” said Thomas Schulthess, Director of CSCS. He highlighted that this sovereign infrastructure fosters open innovation for science and society worldwide. The project also adheres to Swiss data protection and copyright laws.
The release is part of the Swiss AI Initiative, a project launched by EPFL and ETH Zurich in December 2023. The model will be available under the Apache 2.0 license, allowing for broad reuse and further development. EPFL professor Martin Jaggi stated that the open approach aims to drive innovation in Switzerland and Europe and help attract top talent in the field.