Portugal is preparing to launch its first large language model designed specifically for its own language and cultural context, in what Government officials describe as a step towards “digital sovereignty”.
Named ‘Amália’ after the iconic fado singer Amália Rodrigues, the system will go live in early July and is expected to be integrated into public services and selected business applications, as was announced by Prime Minister Luís Montenegro at this year’s Web Summit.
Whilst systems such as ChatGPT and Google’s Gemini have become dominant across the globe, Portuguese officials argue they often perform less accurately in European Portuguese, reflecting their training on predominantly English-language datasets.
The government-backed project aims to address that gap by developing a model trained specifically on Portuguese linguistic and cultural data. Officials say this should improve accuracy, reduce the erroneous blending of European and Brazilian Portuguese, and better capture idioms, historical references, and cultural nuance.
Developers say the model will be capable of generating text, speech, images, and video, as well as summarising documents, answering questions, writing code, and supporting administrative and educational services.
The project brings together around 60 researchers from leading Portuguese universities, including Nova FCT, Instituto Superior Técnico, and the universities of Coimbra, Porto, Minho, Beira Interior, and Évora, alongside the Foundation for Science and Technology.
Rather than being built entirely from scratch, Amália builds on existing European models, including EuroLLM and the Portuguese-focused GlorIA system. Training has been carried out using national and European supercomputing infrastructure, including Deucalion and MareNostrum 5, under the EU’s EuroHPC programme.
The project is funded by around €5.5 million from Portugal’s Recovery and Resilience Plan and has been developed over 18 months, with early beta testing beginning in 2025 before the planned public rollout in 2026. It will also be released as open source, allowing external developers to build on the system.
Government officials say the aim is not to compete directly with commercial AI chatbots, but rather to improve public and business services, where linguistic precision and regulatory compliance are considered particularly important. Early deployment is expected on the gov.pt portal, with wider use across administration and education.
Despite its ambitions, developers acknowledge limitations. Like all AI systems, Amália will be prone to errors and bias, and users will be warned not to treat its outputs as infallible. Safeguards have been introduced to reduce harmful or inaccurate responses, though these are not considered foolproof.
The initiative reflects a wider European push to reduce dependence on US-based AI systems and ensure that data and infrastructure remain under EU regulatory control.
Source: PÚBLICO
Photo credit: Pedro Nunes/Reuters
