Skip to content

Swiss Consortium Launches Apertus, Open-Source Multilingual Language Models

Apertus offers a European, data-regulation compliant alternative. While it shows promise, widespread adoption remains uncertain.

There is an open book on which something is written.
There is an open book on which something is written.

Swiss Consortium Launches Apertus, Open-Source Multilingual Language Models

Apertus, a new family of language models, has been launched by a Swiss consortium. This project aims to provide a transparent, open-source, and multilingual alternative to commercial systems like ChatGPT. The models focus on traceability and support Swiss national languages like Romansh and Swiss German.

The Apertus models were trained using publicly available data, adhering to website robots.txt files and excluding copyrighted, non-permissive, toxic, and personally identifiable content. They are available in sizes with 8 and 70 billion parameters, offering an alternative to systems from US and Chinese tech giants.

The Swiss Bankers Association sees great long-term potential in Apertus, while some banks like UBS already use systems from OpenAI and Microsoft. Industry representatives appreciate the model's compliance with local data protection and banking secrecy laws. Swissmem acknowledges its design for European data regulations, but widespread adoption is not guaranteed.

The Apertus-70B-Instruct model demonstrates solid performance but trails leading open-weight models in most general categories. The models are accessible to researchers, companies, and the public via Hugging Face and can be tested on PublicAI. The project was developed by ETH Zurich, EPFL, and the Swiss National Supercomputing Centre (CSCS) and was publicly released in July 2025. Leandro von Werra of Hugging Face praised Apertus as a 'trailblazer' and 'new milestone in open models' due to its scale and computational intensity.

Apertus, a Swiss-led initiative, offers a transparent, open-source, and multilingual Large Language Model. It aims to foster innovation and build AI expertise while complying with European data regulations. Although it shows promise, widespread adoption remains uncertain.

Read also:

Latest