The Portuguese AI model Amália will have its base version made publicly available in September 2025, according to João Magalhães, coordinator of the project at Nova FCT. Any public or private entity will be able to use the model, provided they have the necessary infrastructure. This release advances previous plans, which had postponed public availability to June 2026.
Amália is a Large Language Model (LLM) focused on European Portuguese and Portuguese culture. It can answer questions, discuss topics, generate code, summarize texts, and explain concepts. The model is being trained on supercomputers MareNostrum 5 and Deucalion, and so far about 25% of the planned development has been completed. The multimodal version—capable of handling text, images, video, and audio—is expected by mid-2026.
The project is fully developed by public Portuguese entities including Nova University of Lisbon, Instituto Superior Técnico, the universities of Porto, Minho, Coimbra, FCT, AMA, and the Institute of Telecommunications, with €5.5 million funding from the Recovery and Resilience Plan. João Magalhães emphasizes that Amália strengthens national technological sovereignty, fosters collaboration among research institutions, and enables applications in education, science, culture, tourism, healthcare, and more.
The September release will allow the academic community, public agencies, and private users to experiment with Amália, while feedback will help improve the model. Early use cases include gov.pt and FCT’s IAedu platform. Magalhães stresses the technical challenges of developing such a model but highlights the strategic importance for Portugal and Europe to lead in the generative AI revolution.
Related News