Switzerland Unveils Apertus, Its First Open-Source National Large Language Model

Public‑Sector Collaboration Builds a National AI Asset

Switzerland’s government‑backed research institutions—EPFL, ETH Zurich and the Swiss National Supercomputing Centre—have joined forces to create Apertus, a large language model positioned as a public‑interest alternative to commercial AI offerings. The effort underscores a vision of AI as essential public infrastructure, comparable to highways, water or electricity. By keeping every aspect of the model’s development open, the partners aim to foster transparency and trust, allowing anyone to scrutinize the training process, source code and underlying datasets.

Model Architecture and Training Scope

Apertus is released in two configurations: an 8 billion‑parameter version and a 70 billion‑parameter version. The model was trained on 15 trillion tokens drawn from more than 1,000 languages, with roughly 40 percent of the data representing non‑English content, including Swiss German and Romansh. Training relied solely on publicly available data, and the crawl process respected machine‑readable opt‑out signals on websites. This approach seeks to align the model with Swiss data‑protection and copyright regulations, providing a compliant option for companies that must adhere to European standards.

Open Access and Broad Usability

The entire Apertus ecosystem—including documentation, source code and the datasets used for training—is publicly released. The model can be accessed via Swisscom, a Swiss ICT provider, or through the Hugging Face platform. This open‑access stance invites a wide range of users—researchers, hobbyists and enterprises—to build customized applications such as chatbots, translation tools, educational aids or other AI‑driven services. By offering a transparent, regulated alternative, Apertus aims to attract sectors that handle sensitive data, especially banking, where compliance with strict local privacy rules is paramount.

Strategic Implications for Swiss Industry

Advocates argue that a domestically developed model can better respect Switzerland’s stringent data‑protection and banking secrecy regulations compared with foreign‑owned AI services. While Swiss banks already employ existing AI solutions, the introduction of Apertus provides a potential homegrown option that could simplify compliance. The model’s public‑infrastructure framing also signals a broader policy direction: treating advanced AI capabilities as a national resource that should be openly available and responsibly governed.

Future Outlook

With Apertus now live, the Swiss AI community is poised to experiment, adapt and extend the model for varied use cases. The open nature of the project encourages collaborative improvement and may serve as a template for other nations seeking to develop sovereign AI capabilities that balance innovation with regulatory compliance.