Switzerland develops its own AI model trained on publicly available data

Switzerland has launched a new open-source AI model, Apertus, which represents an alternative to models such as ChatGPT or Claude. With open source, accessible training data, and detailed documentation, Apertus aims to set new standards for reliable and globally relevant open models.

Apertus (Latin for “open”) is a project designed to provide transparency and accessibility in the world of artificial intelligence. Its code, training data, model weights, and the entire development process are publicly available on the HuggingFace platform. The model was trained on 1,800 languages and is available in two sizes – with 8 billion or 70 billion parameters. It is expected to be comparable in performance to the model Llama 3, which was released by Meta in 2024.

Compliance with European regulations

Apertus was developed with particular emphasis on compliance with European regulatory frameworks. The model is designed to comply with EU copyright law and the voluntary code of conduct for AI. The developers emphasized that the data for training is limited to publicly available sources, while respecting the “opt-out” requirements of websites.

Apertus aims to be more than just another AI model – the project’s ambition is to lay a new foundation for the development of open source solutions that are reliable, ethical and accessible to the global community. At a time when many commercial models are closed and often criticized for their lack of transparency, Apertus opens the door to a more democratic approach to AI development.