ZURICH – In a move set to redefine the landscape of open-source AI, researchers from Switzerland’s leading institutions, ETH Zurich and EPFL, are preparing to release a powerful new large language model (LLM) later this summer. This model is being developed as part of the Swiss AI Initiative and stands out for its complete transparency, multilingual capabilities, and its foundation on public, carbon-neutral infrastructure.
A Commitment to Openness and Transparency
Unlike proprietary models from tech giants, this new LLM will be fully open-source. The project will release not only the model’s weights and code but also its training data under an Apache 2.0 license. This level of transparency is rare and crucial for enabling researchers, developers, and businesses to audit the model for biases, understand its workings, and build on it without proprietary restrictions.
Unprecedented Multilingual Support
A key feature of the model is its massive multilingual reach. The LLM was trained on a diverse dataset of over 1,500 languages, with a focus on providing strong performance in over 1,000 languages. While approximately 60% of the training data is in English, the remaining 40% is non-English, which is a deliberate effort to support under-resourced languages and ensure global inclusivity. This approach makes the model a valuable tool for a wide range of applications, from language preservation to international communication.
Powered by Public, Green Supercomputing
The model was trained on the “Alps” supercomputer at the Swiss National Supercomputing Centre (CSCS). This advanced system, equipped with over 10,000 NVIDIA Grace Hopper Superchips, runs entirely on carbon-neutral electricity. By leveraging public infrastructure, the initiative aims to create a powerful, sovereign AI resource that is not dependent on private corporate interests.
Technical Specifications
The Swiss LLM will be released in two versions to cater to different needs:
- An 8-billion parameter version for lightweight, local use.
- A 70-billion parameter version that is expected to be among the most powerful fully open models globally.
This release represents a significant step towards a more open, transparent, and ethically-aligned AI ecosystem, challenging the closed-source dominance of Big Tech and fostering innovation for the public good.
