What To Know
- Named with the Latin word for “open,” Apertus puts its architecture, training data, code, and model weights fully in the public domain—marking a significant milestone in democratic AI development.
- In the midst of limited access to AI developed by Big Tech, this AI News report reveals how Apertus breaks new ground by including over 1,000 languages—including under-represented tongues like Swiss German and Romansh—and uses 15 trillion tokens for training, with around 40 percent of the data in non-English languages.
AI News: A Bold New Chapter in Sovereign AI
Switzerland has unveiled Apertus, a trailblazing open-source large language model (LLM) developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS). Named with the Latin word for “open,” Apertus puts its architecture, training data, code, and model weights fully in the public domain—marking a significant milestone in democratic AI development.
Swiss researchers unveil Apertus the worlds first fully open AI model
Image Credit: Apertus
Openness at Its Core
Built as a model of transparency and trustworthiness, Apertus is available in two sizes—8 billion and 70 billion parameters—under a permissive open-source licence. Designed for usage in research, education, commercial ventures, and public infrastructure, developers can deploy chatbots, translation tools, and educational applications freely. With access via Hugging Face, Swisscom’s sovereign AI platform, or the international Public AI Inference Utility, it’s accessible to both expert and broader user communities.
In the midst of limited access to AI developed by Big Tech, this AI News report reveals how Apertus breaks new ground by including over 1,000 languages—including under-represented tongues like Swiss German and Romansh—and uses 15 trillion tokens for training, with around 40 percent of the data in non-English languages.
Transparent, Ethical, and Sovereign
Apertus fully respects Swiss and EU norms, including data protection laws and the EU AI Act. Its training process avoids copyrighted or personal data, honors opt-out directives from websites, and applies ethical filters to exclude unwanted material—even refusing to “stealth-crawl.” Documentation and checkpoints are openly available to foster trust and accountability.
Martin Jaggi of EPFL describes this release as a blueprint for “trustworthy, sovereign, and inclusive AI.” Thomas Schulthess of CSCS emphasizes that the project isn’t a simple technology transfer, but a new infrastructure built to nurture Swiss AI expertise across sectors. Imanol Schlag of ETH Zurich highlights that Apertus is “built for the public good,” setting a rare precedent at this scale for multilingual, transparent, and compliant AI.
Hands-On Access and Future Trajectory
During Swiss AI Weeks, developers can experiment with Apertus via Swisscom’s interface, while business clients already have platform access. International users can also tap into the model through the Public AI Inference Utility. This open approach aims to gather usage feedback and foster ongoing model improvements.
Going forward, the Swiss AI Initiative is committed to extending the Apertus family, enhancing efficiency, and crafting domain-specific variants for healthcare, law, climate, and education—while unwaveringly upholding transparency and compliance.
As governments globally wrestle with the twin pressures of AI regulation and innovation, Apertus stands as a bold experiment in public-sector sovereignty and open science. By democratizing access to advanced AI infrastructure and embedding compliance from the outset, Switzerland has injected fresh energy into a conversation that often feels monopolized by private giants.
The Swiss model challenges prevailing norms—suggesting that powerful AI need not be walled behind paywalls or commercial interests. Instead, AI can serve as digital infrastructure for all—if built with transparency, inclusivity, and civic values at heart. For Thailand and ASEAN, Apertus offers both inspiration and a potential reference point for developing sovereign, ethical, and locally relevant AI systems that respect linguistic diversity and public needs.
For more details on Apertus, visit:
https://www.swiss-ai.org/apertus
For the latest on Apertus, keep on logging to Thailand AI News.
