AMÁLIA — Portugal's Open-Source Large Language Model for European Portuguese

Portugal sought sovereign AI infrastructure for European Portuguese amid a broader European debate on AI sovereignty.

AMÁLIA is Portugal's first large language model built specifically for European Portuguese, intended as open-source public digital-sovereignty infrastructure.

Developed by a five-university consortium — coordinated by NOVA University Lisbon with Instituto Superior Técnico and the universities of Coimbra, Porto and Minho, involving more than 60 researchers — the model extends the open, European-funded EuroLLM-9B foundation model to 9 billion parameters, plus separate vision and speech-recognition components and a 32,000-token context window. It was released open-source under an Apache 2.0 licence on Hugging Face, funded through Portugal's Recovery and Resilience Plan (PRR).

Publicly unveiled 1 July 2026, AMÁLIA has no public chat interface and experimental search capability only; it is running in a limited beta with four validated pilot users (museum, science, media and education bodies) for tasks such as answering visitor questions about artworks. The project cost around EUR 7 million through 2027, with a roadmap toward a 22-billion-parameter version by 2027.

Independent Portuguese and international press explicitly caution that the model 'is not yet fully operational,' that it is an adaptation of an existing multilingual model rather than an original build, and that at 9 billion parameters it is far smaller than leading commercial systems — an honest, evidence-thin starting point rather than a finished public-sector deployment.