This summer, the École Polytechnique Fédérale de Lausanne (EPFL) and ETH Zurich will jointly release a large language model (LLM) trained on public infrastructure. The model was trained on the Alps supercomputer at the Swiss National Supercomputing Centre (CSCS), marking a significant milestone in open-source artificial intelligence and multilingual excellence.

Recently, around 50 leading global open-source LLM and trustworthy AI initiatives gathered in Geneva for the International Open-Source LLM Builders Summit. Hosted by the AI centers of EPFL and ETH Zurich, the summit represented a key step in building an international ecosystem for open foundation models. The event previewed the upcoming open-source LLM, developed collaboratively by researchers from EPFL, ETH Zurich, other Swiss universities, and CSCS engineers. Currently in final testing, the model will be available for download under an open license, emphasizing transparency, multilingual performance, and broad accessibility.

The model will be fully open, with source code, weights, and training data publicly released, supporting adoption in science, government, education, and the private sector. ETH Zurich AI Center research scientist Imanol Schlag stated: "Fully open models enable high-trust applications and are crucial for advancing research on AI risks and opportunities." The model covers over 1,000 languages, trained on a large text dataset including more than 1,500 languages, ensuring global applicability. It will be released in 8 billion and 70 billion parameter versions to meet different user needs, with the 70 billion version becoming one of the world's most powerful open-source models. Training used over 15 trillion high-quality tokens, ensuring high reliability and strong language understanding capabilities. Development strictly complies with Swiss data protection laws, copyright laws, and transparency obligations under the EU AI Act. The model was trained on the Alps supercomputer at CSCS in Lugano, equipped with over 10,000 NVIDIA Grace Hopper Superchips and powered by 100% carbon-neutral electricity. CSCS Director Thomas Schulthess said: "Strategic investment in a supercomputer purpose-built for AI has enabled us to train such models." By late summer, the LLM will be released under the Apache 2.0 license, accompanied by detailed documentation to promote transparent reuse and further development. EPFL Professor Jaggi stated: "Through full openness, we hope to drive innovation in Switzerland, Europe, and multinational collaborations, attracting top talent."