October 1, 2026, (Inside AI) — Two fully open large language models built by European public institutions are now available on a major commercial AI inference platform, marking a significant test of whether sovereign AI can scale beyond government pilot programs. EuroLLM, which covers all 24 official European Union languages, and Apertus, Switzerland's first fully open multilingual model trained on more than 1,500 languages, can be accessed through Cloudflare's Workers AI platform starting today, according to the company.
The move arrives exactly one year after Cloudflare first argued that AI sovereignty should be defined by choice rather than control. That argument has since been tested by a hardening geopolitical landscape. Attackers have used frontier models to conduct cyber operations. Access to some frontier models now depends on geographic location. Calls to restrict open model development have grown louder in Washington and Brussels alike.
EuroLLM was developed by a consortium including Instituto Superior Técnico, the University of Edinburgh, Université Paris-Saclay, Sorbonne University, Unbabel, Naver Labs and the University of Amsterdam. It received support from Horizon Europe, the European Research Council and EuroHPC. The model was trained on the MareNostrum 5 supercomputer in Barcelona and supports 35 languages total. According to the consortium, it outperforms similar-sized models on EU multilingual benchmarks and machine translation tasks.
Apertus, Latin for "open," was built by ETH Zurich, EPFL and the Swiss National Supercomputing Centre as part of the Swiss AI Initiative. It was trained on more than 15 trillion tokens, with 40% of training data in languages other than English. Its architecture, weights, training data and methods are all published publicly. The model was designed with Swiss and European regulations in mind, including the EU AI Act and GDPR, which means respecting training opt-outs, removing personal data and preventing memorization of sensitive information. It was trained on CSCS's Alps supercomputer, which contains more than 10,000 GH200 GPUs.
Read: As AI Accelerates, Governments Are Increasingly Left Behind
"Its developers report that it significantly outperforms leading closed and open models on rare and regional languages, from Romansh and Swiss German to low-resource languages across Asia and Africa," according to documentation provided by the model's developers.
Asia-Pacific Builders Show Open Models Work
The practical case for open models has been building across Asia-Pacific. Since Cloudflare added national models from India, Japan and Singapore to its platform last year, hundreds of students, startups and public servants have built applications on them. Three examples illustrate the pattern.
In India, students at the Indian Institute of Technology Delhi built Form Mitra, a voice-guided assistant that walks citizens through government benefit forms in any of 22 Indian languages. The tool addresses a specific failure: forms written in dense English shut out many of the rural, low-literacy and visually impaired citizens they are meant to serve. It uses AI4Bharat's IndicTrans2 model and was developed at a buildathon run with CyberPeace.
In Singapore, MedBridge guides patients through health conversations in 14 languages, including Hokkien and Cantonese. The tool addresses a clinical communication gap: many nurses caring for Singapore's elderly patients come from across Southeast Asia and do not speak local dialects, so critical information can get lost in translation.
In Japan, Anshin Concierge lets elderly residents and people with disabilities describe problems in their own words, such as "my knees hurt" or "a strange screen appeared on my phone," and connects them to the right Tokyo Metropolitan Government support desk by phone or web page in one tap. It began as a hackathon prototype built with Code for Japan.
These applications share a common thread. They serve populations that commercial AI models often overlook because the languages are low-resource or the users are not profitable enough to prioritize.
Read: AI Agents Breach Government Systems in US and Australia, Sparking Calls for Oversight
Security Teams Confront Model Dependency
The sovereignty argument has also moved into cybersecurity. Governments want to use AI to defend essential services and national infrastructure. Frontier models that can find vulnerabilities at scale can find them for defenders too. But a defense built on one model is only as dependable as access to that model, and governments have watched access to critical models get constrained with little warning.
Cloudflare's security team confronted the same problem internally. Over the past year, the company built what it calls a harness: an orchestration layer that coordinates multiple AI models working in parallel to hunt for vulnerabilities, verify findings and prioritize threats. The company open-sourced the harness so any organization can run it with the models of their choice.
"Because the harness works with any model, closed or open, losing access to one provider doesn't switch your defenses off," according to company documentation.
When Cloudflare walked governments through the system, the most common reaction was relief, followed by practical questions about how to deploy it in their own environments under their own rules. Briefings quickly turned into requests for hands-on training.
Starting this month, Cloudflare is launching a program of hands-on workshops for government cybersecurity agencies and critical infrastructure operators. Participants build their own AI security harness and layered defenses, and leave knowing how to adapt both to their organization. The modules are plug-and-play, designed to slot into national AI skilling and cyber resilience programs. The first workshop runs in Singapore this October at Singapore International Cyber Week.
The timing reflects a broader shift. In February, Cloudflare CEO Matthew Prince told the India AI Impact Summit 2026 in New Delhi that decentralized, affordable access to AI is a matter of national resilience. The summit series moves to Geneva in June 2027, with a mission of "prosperity and progress for all."
Cloudflare's network runs in more than 335 cities across 125+ countries, with GPUs for AI inference in more than 230 of them. The company says it does not think any country should have to depend on one company for its AI, including itself.
Organizations interested in the workshops or the open models can request access to EuroLLM and Apertus on Workers AI, or contact Cloudflare's policy team directly. The company says the past year demonstrated that choice is not just a path to AI sovereignty but also to AI security.
What remains unclear is whether open models built by public institutions can sustain development at the pace of commercial frontier labs, which spend billions annually on compute and talent. The European and Swiss models represent a significant investment, but they compete against systems with far larger budgets. The answer may determine whether sovereign AI remains a viable alternative or becomes a niche for specialized applications.
