Switzerland has unveiled Apertus, a fully open large language model (LLM) designed as a transparent alternative to proprietary systems like ChatGPT.
The model, released by EPFL, ETH Zurich, and the National Supercomputing Centre, provides complete access to its source code, model weights, and training data.
Trained on 15 trillion tokens across more than 1,000 languages—including Swiss German and Romansh—Apertus emphasizes multilingual diversity. It comes in two scalable versions: 8 billion and 70 billion parameters, making it suitable for both research and large-scale applications.
What sets Apertus apart is its compliance-first approach. It was trained only on publicly available datasets, respects website crawler opt-outs, removes personal data, and aligns with EU copyright rules and the upcoming EU AI Act.
The model is available via Hugging Face for direct download and through Swisscom’s AI Platform for cloud-based use. With this release, Switzerland aims to strengthen digital sovereignty, promote ethical AI development, and offer researchers and businesses a powerful, transparent alternative to closed systems.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
