EuroLLM: LLM made in Europe built to support all 24 official EU languages (eurollm.io)

EuroLLM is an open-source large language model initiative from a European research consortium that delivers native support for all 24 official EU languages. Its flagship EuroLLM-9B is a 9-billion-parameter model pretrained on over 4 trillion tokens across 35 languages (covering every EU language) and available as a Base model for fine-tuning plus an Instruct variant tuned for chat and instruction following. A lighter EuroLLM-1.7B is also released for edge use. Models were trained on the MareNostrum 5 supercomputer with support from Horizon Europe, the ERC and EuroHPC, and the project claims superior performance against similar-sized alternatives on tasks such as question answering, summarization and translation. All assets are freely available on Hugging Face. The project matters for AI/ML because it prioritizes comprehensive multilingual coverage, transparency and European digital sovereignty—helping close gaps for low-resource EU languages while enabling researchers and companies to fine-tune and deploy models locally. Technically notable points: multi-size checkpoints for edge and server workloads, an instruction-tuned chat variant, training scale (4T tokens) and HPC-backed compute, and an explicit roadmap toward multimodality (vision and voice). By combining open licensing, competitive performance and institutional backing, EuroLLM aims to be a reusable foundation for multilingual NLP research, productization and regulatory-aligned deployments across Europe.