THE CRUNCH
Cloudflare has added two new open-source language models to its Workers AI platform, expanding the range of sovereign AI tools available to developers and public institutions. EuroLLM, a European Union-focused model supporting 35 languages, and Apertus, a Swiss model trained on more than 1,500 languages, are now available for access. The move follows a year of growing debate over AI access and sovereignty, with Cloudflare arguing that open models offer a resilient alternative to closed systems.
EuroLLM was developed by a consortium of European universities and research institutions, including the University of Edinburgh and Naver Labs, and was trained on the MareNostrum 5 supercomputer. The model is designed to outperform similar-sized models on EU multilingual benchmarks and machine translation, with a focus on underserved languages. Apertus, developed by ETH Zurich, EPFL and the Swiss National Supercomputing Centre, is a fully open model with published weights and training data. It was trained on the Alps supercomputer and is designed to respect EU AI Act and GDPR rules, including training opt-outs and data removal.
Cloudflare reports that since adding national models to Workers AI last year, hundreds of students, startups and public servants across Asia-Pacific have built applications on them. Examples include Form Mitra, a voice-guided assistant that helps rural citizens fill out government forms in their native language, and MedBridge, a tool that guides patients through health conversations in 14 languages. The company also announced workshops for government cyber agencies to build AI defenses that work with any model, with the first event scheduled for Singapore in October.