Europe takes another step toward AI sovereignty with the launch of EuroLLM-22B

A European Consortium including the University of Edinburgh has launched EuroLLM-22B, a fully open LLM built for all 24 EU languages, setting a new benchmark for multilingual AI backed by EU-made supercomputing power.

Alexandra Birch and the Instituto de Telecomunicações

EuroLLM-22B has been developed by a consortium of leading academic and industrial partners, including the University of Edinburgh, Paris-Saclay University, the University of Amsterdam, Unbabel and Naver Labs.

EuroLLM-22B is the largest model in the EuroLLM series, following EuroLLM-1.7B and EuroLLM-9B. Trained from scratch on the MareNostrum 5 supercomputer at the Barcelona Supercomputing Center, it delivers competitive - or superior - performance to global models of similar size on multilingual tasks, setting a new standard for multilingual LLMs and marking an important milestone for Europe’s mission to accelerate “EU-made” AI innovation.

Europe’s uniquely strong public supercomputing network, coordinated by the EuroHPC Joint Undertaking (EuroHPC JU), enabled the large-scale training effort. With three machines ranked among the world’s Top 10 and another exascale system launching soon, EuroHPC demonstrates how European infrastructure can power state-of-the-art AI development globally.

Open Access

A defining feature of EuroLLM-22B is its fully open nature. The model is freely available to researchers, startups, developers, and public institutions, lowering barriers to entry and enabling experimentation across Europe’s innovation ecosystem.

We set out to create a model that would be a genuine resource for the European research community and beyond. At 22 billion parameters, this model strikes an important balance, powerful enough for serious applications, yet accessible enough for academic institutions and smaller organisations to deploy and study.

We’re very proud to release EuroLLM-22B today. This has been a long journey and an incredible team work across many levels, including data quality filtering, pretraining, long-context adaptation, and post-training. We see this as another exciting and impactful step to strengthening Europe’s sovereignty in AI. Our goal is that EuroLLM becomes a flywheel for innovation so that anyone can develop on top of it. EuroLLM is a success story for the European supercomputing network and how it can help advance AI—proof that amazing things can happen through open collaboration.

With major players like OpenAI, Google, Alibaba, and Meta dominating the AI landscape, reliance on their models poses significant risks, including limited openness and uncertain availability. Together with other European initiatives, such as OpenEuroLLM and Apertus, EuroLLM aims to counter this trend by offering a fully accessible alternative designed to serve Europe’s needs without compromising its independence.

EuroLLM emerged from a pressing need to bridge gaps in language access across the EU, and create a model tailored to Europe’s linguistic and cultural diversity. Its comprehensive language coverage and the potential of this model to drive inclusive innovation - from public services to private enterprise - is at the heart of its premise. While its initial focus is multilinguality - supporting all 24 official EU languages as well as 11 additional languages - EuroLLM will extend the model into multimodal capabilities - including speech, vision, and video - powered by a 2M GPU-hour EuroHPC extreme-scale grant on the Jupiter exascale system starting in 2026.

EuroLLM-22B signals a turning point for Europe’s role in shaping the future of AI. By uniting world-class researchers, leveraging a European-wide supercomputing network, and committing to full openness, the EuroLLM initiative shows how Europe can lead and create impact with values of accessibility, diversity, and scientific excellence.

EuroLLM-22B is available today via Hugging Face, where technical documentation and benchmark comparisons are publicly accessible.

Open Access

Related links