EuroLLM 9B

EuroLLM 9B is a 9-billion-parameter open language model built by a nine-institution European consortium led by Unbabel, including Instituto Superior Tecnico, the University of Edinburgh, and the University of Amsterdam. Trained on 4 trillion tokens on the MareNostrum5 supercomputer, it supports 35 languages, covering all 24 official EU languages plus 11 additional commercially and strategically important languages.

Development was funded with 20.6 million euros from the EU Digital Europe programme. Alongside the base and instruct-tuned variants, the project openly released its multilingual data filter EuroFilter and the EuroBlocks-Synthetic post-training dataset, making it one of the most transparent European model efforts to date. Check the Hugging Face model card for the exact license terms.