Switzerland has officially entered the global artificial intelligence arena with the launch of its new large language model (LLM), Apertus. Developed as a major national initiative led by prestigious institutions EPFL and ETH Zurich, in collaboration with the Swiss National Supercomputing Centre, this project represents a significant collaborative effort. It brings together a wide range of engineers, researchers, and students from across the country, spanning multiple disciplines.
The model's name, Apertus, is derived from the Latin word for ‘open,’ which is a core principle guiding its entire development philosophy. This marks a distinct approach in a field currently dominated by proprietary models from major tech companies.
A Multilingual and Inclusive Approach
A key differentiator for Apertus is its extensive and inclusive training data. The model has been trained on a massive dataset of 15 trillion tokens encompassing more than 1,000 languages. The developers placed a specific emphasis on incorporating languages that have been historically underrepresented in existing LLMs. This includes Swiss German and Romansh, a move that aims to preserve and empower local linguistic diversity within the digital AI landscape.
The Hallmark of Transparency
Positioned as a transparent alternative to models like Meta’s Llama and OpenAI’s ChatGPT, Apertus is defined by its commitment to openness at every stage. The developers have made its architecture, model weights, training data, and training recipes "openly accessible and fully documented." This level of transparency is unprecedented and allows for greater external scrutiny, collaboration, and trust.
Adherence to Strict Ethical and Legal Standards
The development team stated that Apertus was created with careful consideration for Swiss data protection laws, copyright laws, and the transparency obligations outlined in the EU AI Act. A significant focus was placed on data integrity and ethical standards from the outset.
The training corpus is built exclusively on publicly available data that has been rigorously filtered. This process respects machine-readable opt-out requests from websites—even applying them retroactively—and removes personal data and other undesired content before the training process begins. This proactive approach to data sourcing is a direct response to growing concerns in the industry.
A Response to Industry-Wide Concerns
The launch of Apertus comes at a critical time when AI developers face intense criticism over their data gathering, deployment, and storage practices. The industry has been rocked by lawsuits, such as the one filed by a group of authors against Microsoft in June 2025 for allegedly using copyrighted works to train its models.
Similarly, significant privacy concerns were raised in 2024 when Elon Musk’s AI chatbot, Grok, was found to be using X's user data for training without obtaining explicit consent. The Swiss initiative, with its foundational principles of transparency and ethics, presents a compelling alternative model for responsible AI development that prioritizes user rights and legal compliance.
