Switzerland Launches Open-Source AI Model Apertus Trained on Public Data with EU Compliance

Switzerland has unveiled its own open-source AI model named Apertus, launched on Monday as a publicly available alternative to proprietary systems like OpenAI’s ChatGPT or Anthropic’s Claude. The model, whose name means “open” in Latin, is now accessible on the Hugging Face platform, where its source code, training data, model weights, and full development process are openly shared. Developed by a team of researchers and institutions in Switzerland, Apertus aims to establish a new benchmark for trustworthy, transparent, and globally relevant open AI models. The model was trained on content from over 1,800 languages and is available in two versions: one with 8 billion parameters and another with 70 billion parameters—making it comparable in scale to Meta’s 2024 Llama 3 release. A key focus of Apertus is compliance with European Union regulations, particularly the EU’s AI Act and copyright laws. The developers emphasized that the model’s training data was sourced exclusively from public content, with strict adherence to website opt-out requests from AI crawlers. This means no unauthorized or “stealth-crawling” of web content, reinforcing ethical data collection practices. The initiative reflects Switzerland’s broader commitment to responsible AI development, positioning Apertus as a model that balances innovation with legal and ethical accountability. While some U.S.-based AI companies have expressed concerns that EU regulations may slow down AI progress, the Apertus project demonstrates that robust compliance and openness are not mutually exclusive with high performance. By making the model fully open and transparent, the Swiss team hopes to foster global collaboration, encourage scrutiny, and provide a reliable foundation for researchers, developers, and organizations seeking alternatives to closed, proprietary AI systems.