Barcelona Supercomputing Center

Maker of Spain's publicly funded ALIA and Salamandra models.

The Language Technologies Lab at the Barcelona Supercomputing Center builds Spain's public language models, trained from scratch on MareNostrum 5, the EuroHPC pre-exascale machine it hosts and operates. The ALIA family is pretrained on 9.83 trillion tokens spanning 35 European languages, with post-training concentrated on Spanish, Catalan, Basque, Galician, and English, and released under Apache 2.0 with the training scripts and configurations published alongside the weights. The work is funded by Spain's Ministry for Digital Transformation and supported by the Catalan government's Aina project, which makes ALIA one of the few frontier-scale models built as public infrastructure rather than a commercial product.