Switzerland has released Apertus, a large language model (LLM) built entirely on open principles, making every aspect of its design, training data, and documentation publicly accessible. Developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS), Apertus is designed to serve as a foundation for research, education, and commercial applications.
The name Apertus—Latin for “open”—reflects the project’s core philosophy: AI developed transparently, inclusively, and for the public good.
A Truly Open AI Model
Unlike many commercial AI systems that disclose only parts of their training process, Apertus is being released under a permissive open-source license. Both the 8-billion-parameter and the 70-billion-parameter versions are freely available on Hugging Face or via Swisscom, the project’s strategic partner.
“With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed,” said Martin Jaggi, Professor of Machine Learning at EPFL and member of the Swiss AI Initiative Steering Committee.
The model will be updated regularly by teams at EPFL, ETH Zurich, and CSCS, ensuring Apertus evolves alongside advances in AI research.
Built for Multilingual Reach
Apertus was trained on 15 trillion tokens across more than 1,000 languages, with roughly 40% of the data coming from non-English sources. This includes underrepresented languages such as Swiss German and Romansh, making Apertus one of the few large-scale multilingual open models available today.
“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 principles,” said Imanol Schlag, Technical Lead and Research Scientist at ETH Zurich.
Real-World Use and Access
Apertus is already being deployed by Swisscom on its sovereign AI platform.
“This underscores our commitment to shaping a secure and responsible AI ecosystem that serves the public interest and strengthens Switzerland’s digital sovereignty,” said Daniel Dobos, Research Director at Swisscom.
For developers, Apertus can be tested during the ongoing Swiss {ai} Weeks until October 5, 2025. Participants in hackathons will gain access via Swisscom-hosted interfaces, while international users can experiment with the model through the Public AI Inference Utility.
Joshua Tan, Lead Maintainer of the utility, framed the project as public infrastructure: “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.”
Compliance and Transparency
The Apertus training process followed Swiss data protection laws, copyright rules, and the EU AI Act’s transparency requirements. Training data was sourced only from publicly available materials, filtered to exclude personal data, honor website opt-out requests, and remove sensitive or harmful content.
This emphasis on compliance, the developers argue, makes Apertus a trustworthy benchmark in a landscape often criticized for secrecy and opaque data sourcing.
The Road Ahead
The Swiss AI Initiative sees Apertus as just the beginning. Future plans include expanding the Apertus family, improving efficiency, and developing domain-specific models tailored to law, healthcare, climate science, and education—all while upholding strict standards of openness and accountability.
“Apertus demonstrates that generative AI can be both powerful and open,” said Antoine Bosselut, Professor at EPFL and Co-Lead of the Swiss AI Initiative. “The release of Apertus is not a final step, but the beginning of a long-term commitment to open, trustworthy, and sovereign AI foundations, for the public good worldwide.”
