Key Highlights-
- Switzerland has unveiled Apertus, a new, open-source AI language model trained exclusively on publicly available data.
- The model is developed by a consortium of Swiss institutions, and offers a “blueprint” for transparency, complying with European data and copyright laws.
- Interested users can download a copy of Apertus onto their own servers to build projects.
Switzerland has officially launched Apertus, a new open-source national Large Language Model(LLM), resulting in a direct challenge to proprietary AI chatbots. Developed by a joint collaboration between EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS), Apertus AI stands out for its ‘radical transparency.’
To keep things fair, the AI’s entire development process has been documented, from its architecture and training data to its model weights and it openly accessible on its Hugging Face page.
Apertus is designed with transparency at its core, thereby ensuring full reproducibility of the training process.
-Swiss AI Initiative
This approach sets it apart from its commercial counterparts, and positions itself as a public-interest alternative to the currently closed-source models in the AI market.
Training Apertus on Public Data & Open-Source Ethos
Apertus, whose name is Latin for “open,” represents a clear stance on the future of AI. Unlike for-profit models like ChatGPT, Grok or Gemini to name a few, Apertus was trained exclusively on publicly available information keeping in mind European data protection and copyright laws. The developers went a step even further by implementing a system to honour “machine readable optout requests” from websites, filtering out personal data in the process. This commitment to privacy and compliance makes Apertus a trustworthy alternative, given how other models have faced legal challenges over data scraping.
As a “fully open language model,” Apertus allows researchers, professionals, and enthusiasts alike to inspect and build upon the model itself. According to Martin Jaggi, a professor at EPL and a member of the Swiss AI Initiative, this openness is a key feature. “With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed,” Jaggi said.
The multilingual model designed to cater linguistic diversity, with 40% of its training data being non-English, including languages like Swiss German and Romansh, making it especially useful within its native country.
Apertus is available to download for free in 8 billion and 70 billion parameters under the Apache 2.0 license. You can try out the chat interface at PublicAI’s official website.
The Sovereign Alternative to Commercial AI Giants
Most proprietary models like ChatGPT and Gemini are currently leading in raw power or meeting consumer’s expectations, Apertus offers its own unique proposition. The model would lack on a feature-to-feature basis but on the basis of trust, sovereignty and trust, it would perform well – all areas where commercial counterparts are most vulnerable.
Operating on public infrastructure principles, the Swiss AI, does not harvest data for advertising, has no hidden surveillance, and complete transparency about how it works. According to Swissinfo.ch’s report the Swiss Bankers Association that the home-grown LLM has a promising long-term potential, given its compliance with the data protection and banking secrecy laws. Even with its open stance, the AI may not be the first choice for all users. Other Swiss banks including the largest bank in Switzerland – UBS, has already tied up with AI giants like OpenAI and Microsoft. Adding to this, Swissmem’s head of digitalisation, innovation and technology, Adam Gontarz said –
Experience shows that there is no single solution that fits all needs….Each project must take into account the specific circumstances and requirements of the business. In some cases, international solutions may also be the most effective choice.
The Apertus team is determined to explore domain-specific areas like healthcare, climate science, and legal research in the future iterations. Recently, xAI owner Elon Musk also open-sourced Grok 2.5 with Grok 3 following soon after, with similar intent of transparency and winning users trust.
