Portugal Unveils Amália: Europe’s First Open‑Source Portuguese LLM

On 1 May, the Portuguese government announced the official release of Amália, the first open‑source large language model (LLM) developed specifically for European Portuguese.

Amália is a national‑strategic project involving more than 60 researchers from several Portuguese academic institutions. Development took 18 months and was financed with an initial €5.5 million investment, leveraging the Deucalion and MareNostrum 5 super‑computing infrastructures.

The model builds on the EuroLLM‑9B foundation model. In its first stage, it was trained on approximately 4 × 10¹³ Portuguese words, resulting in a 9 billion‑parameter LLM that now supports text, image, and audio inputs, an expanded context window, enhanced safety mechanisms, and a dedicated evaluation system.

Initial deployments will target Portuguese government and public‑sector services, including education, defense, health, culture, and citizen‑service applications such as AI teaching assistants, virtual museum guides, digital citizen‑service bots, and decision‑support tools for the navy.

The project is funded through Portugal’s Recovery and Resilience Plan (PRR), with the €5.5 million grant distributed to Lisbon University, Lisbon Technical Institute, University of Porto, University of Minho, University of Coimbra, and the Portuguese Science and Technology Foundation (FCT). An additional €1.5 million has been secured for the next phase (through 2027), and a larger 22 billion‑parameter version with autonomous agent capabilities is planned for release later this year.

The test version was completed in September 2025 and demonstrated at the PROPOR conference in Brazil.