Switzerland Launches Fully Open-Source AI Model

A consortium of leading Swiss institutions has unveiled Apertus, a groundbreaking open-source AI model designed to serve as a foundational platform for future research and diverse applications. Developed through a collaborative effort by EPFL (École polytechnique fédérale de Lausanne), ETH Zurich (Swiss Federal Institute of Technology Zurich), and the Swiss National Supercomputing Centre (CSCS), Apertus, meaning “open” in Latin, embodies its core principle: complete transparency in design, training data, and development processes.

This significant move challenges the closed-door approach often associated with proprietary AI models, potentially democratizing access to advanced AI technology. Unlike systems that often shroud their inner workings in secrecy, Apertus offers full access to its architecture, training methodologies, and comprehensive documentation. This level of transparency is crucial for fostering trust and enabling researchers and developers worldwide to understand, scrutinize, and build upon the model’s capabilities.

Apertus is positioned to empower developers and organizations to create a wide range of applications, from sophisticated chatbots and translation services to specialized educational tools. Available in two versions – an 8-billion-parameter variant and a larger, more powerful 70-billion-parameter model – Apertus can be readily downloaded or accessed through Swisscom, the initiative’s strategic partner. Both versions are released under a permissive open-source license, facilitating their widespread adoption in research, education, and even commercial projects.

Martin Jaggi, Professor of Machine Learning at EPFL and a member of the Steering Committee of the Swiss AI Initiative, emphasizes the project’s commitment to responsible AI development: “With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed.” Jaggi added that Apertus will undergo regular updates by a dedicated team of engineers and researchers from CSCS, ETH Zurich, and EPFL. This commitment to ongoing improvement and maintenance is critical for ensuring the model’s long-term viability and continued relevance.

Thomas Schulthess, Director of CSCS and Professor at ETH Zurich, envisions Apertus as “a driver of innovation and a means of strengthening AI expertise in research, society and industry.” He distinguishes this project from a mere technology transfer, emphasizing its aim to establish a robust infrastructure designed for sustained use and contribution to the global AI landscape.

One of the distinctive features of Apertus is its multilingual reach. The training process involved an impressive 15 trillion tokens spanning over 1,000 languages, with approximately 40% of the data originating from non-English sources. Notably, Apertus includes support for languages often neglected in larger language models (LLMs), such as Swiss German and Romansh. This localized focus could give the model an edge in nuanced, regional language processing tasks that generic LLMs might struggle with.

Imanol Schlag, technical lead of the project and Research Scientist at ETH Zurich, frames Apertus as a public good: “Apertus is built for the public good. It stands among the few fully open LLMs at this scale and is the first of its kind to embody multilingualism, transparency, and compliance as foundational design principles.” This commitment to transparency extends to the ethical considerations incorporated into its development. The training dataset was carefully curated using only publicly available data, rigorously filtered to remove personal information and respect website opt-out requests. Ethical guidelines were also applied to proactively exclude undesirable material before training commenced.

Swisscom has already begun deploying Apertus on its sovereign AI platform, underscoring the model’s practical utility. Daniel Dobos, Research Director at Swisscom, states, “This underscores our commitment to shaping a secure and responsible AI ecosystem that serves the public interest and strengthens Switzerland’s digital sovereignty.” This move by Swisscom highlights the growing importance of national AI strategies and the desire for greater control over AI infrastructure.

Joshua Tan, Lead Maintainer of the Public AI Inference Utility, argues for the broader significance of Apertus: “Currently, Apertus is the leading public AI model: a model built by public institutions, for the public interest. It is our best proof yet that AI can be a form of public infrastructure like highways, water, or electricity.” This analogy positions Apertus and similar open-source initiatives as vital components of a modern digital infrastructure, accessible and beneficial to all.

Future iterations of Apertus are planned to expand the model family, enhance efficiency, and develop domain-specific tools tailored for sectors such as law, healthcare, climate science, and education. These future development efforts will consistently prioritize the values of transparency, trustworthiness, and ethical compliance. The emergence of Apertus represents a significant step towards a more open, collaborative, and accountable AI future, signaling a potential shift in the power dynamics of the AI industry.