A coalition of Swiss institutions—including EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS)—has launched Apertus, an open AI model designed with accessibility at its core. Every detail, from its architecture to its training data, is available for public inspection. The project aims to set a new standard for trustworthy and sovereign AI, offering a blueprint for how open research can benefit society at large.
Accessibility and applications
Apertus comes in two versions: an 8-billion-parameter model and a larger 70-billion-parameter model. Both are available under a permissive open-source licence, making them usable across education, research, and commercial projects. Developers can download the models directly from Hugging Face or access them through Swisscom, which is already deploying Apertus on its sovereign AI platform.
Multilingual by design
Unlike many large language models that remain English-centric, Apertus was trained on 15 trillion tokens across more than 1,000 languages, with 40% of the data in non-English sources. This includes languages often excluded from AI development, such as Swiss German and Romansh. The multilingual foundation reflects the project’s goal of inclusivity and cultural representation.
Public infrastructure approach
Project leaders describe Apertus as more than just another AI model. They frame it as a form of digital public infrastructure—comparable to utilities like electricity or highways—built by public institutions to serve collective interests. Researchers emphasize that Apertus will be continuously updated, providing a long-term foundation for innovation in fields such as healthcare, climate, education, and law.
Compliance and ethics
The model’s development followed strict guidelines under Swiss data protection and copyright rules, while also aligning with the EU AI Act’s transparency standards. Training relied solely on publicly available datasets, with steps taken to exclude personal information and ethically sensitive content. This compliance-first approach is intended to build trust and encourage safe adoption across industries.
Looking ahead
For now, Apertus is available for experimentation through Swisscom’s AI platform and via Hugging Face, with hands-on access being offered at Swiss {ai} Weeks hackathons. Future plans include expanding the model family, improving efficiency, and creating domain-specific tools. The initiative underscores Switzerland’s ambition to position itself as a leader in responsible, open, and multilingual AI innovation.
Source: https://www.artificialintelligence-news.com/news/switzerland-releases-its-own-fully-open-ai-model/
