CLOUDFLARE marks a year since its Sovereign AI initiative by highlighting two European open models now accessible on Workers AI: EuroLLM and Apertus. EuroLLM covers 35 languages, including all 24 official EU languages, and was developed with Horizon Europe, the European Research Council and EuroHPC by a consortium that includes academic and research partners such as the University of Edinburgh and Unbabel.
Apertus, Switzerland’s first fully open multilingual model, was trained on more than 15 trillion tokens across over 1,500 languages, with 40% of training data in languages other than English. Both are designed to be used openly, with access requests available to researchers and public sector users. Apertus’ design aligns with Swiss and European norms, including EU AI Act and GDPR, and its training process emphasises opt-outs and minimising memorisation.
Training utilised CSCS’s Alps supercomputer (over 10,000 GH200 GPUs). Both models are presented as alternatives to vendor-locked approaches and are offered via Workers AI for governments, startups and civil society to experiment with.
Beyond model access, Cloudflare describes practical AI defence work that remains effective even if a single model becomes unavailable. The company has built an orchestration harness that runs multiple models in parallel to hunt for vulnerabilities, verify findings and prioritise threats; it has open-sourced this harness and published a layered architecture for frontier and open models to work together.
A programme of hands-on workshops for government cybersecurity agencies and critical infrastructure operators is being launched, with the first session in Singapore in October. The article also cites real-world builds by APAC participants, including Form Mitra in India, MedBridge in Singapore and Anshin Concierge in Japan, illustrating how open models can support accessible public services. Cloudflare argues that choice enhances AI resilience and security, not just sovereignty.