Fernando Alexandre Confirms Amália Portuguese Language Model for Public Release by End-June at the Macau Stop — €5 Million PRR Build-Out Closes the Two-Year Cycle

The Minister for Education, Science and Innovation, Fernando Alexandre, confirmed on Monday from Macau that the Portuguese large language model Amália will be made public “this month” — closing a two-year build cycle the State priced at more than €5 million and routed almost entirely through the Plano de Recuperação e Resiliência (Recovery and Resilience Plan, PRR). The model has been in restricted research-only access since the April 2025 beta freeze and now moves toward an open licence framework the minister has previously described as “usable by any entity, including private companies”.

Amália sits inside Portugal’s Agenda Nacional de Inteligência Artificial (National AI Agenda) and was developed by a consortium led by the Universidade Nova de Lisboa, the Instituto Superior Técnico, the Universidade de Coimbra, the Universidade do Porto, the Universidade do Minho and the Fundação para a Ciência e a Tecnologia (Foundation for Science and Technology, FCT). The project is built specifically around European Portuguese — explicitly not a fork or fine-tune of an English-first base — and aims, in Alexandre’s framing, to give “greater rigour in all dimensions of Portuguese language and culture” while opening a path for adoption across the Lusophone bloc.

The minister’s confirmation in Macau is the first hard public-date commitment since a March pre-announcement that pencilled in June for the full release. Government has so far ruled out a paid tier: the model will not have a freemium-style consumer wrapper, but a fully redistributable artefact that universities, start-ups and the Administração Pública (Public Administration) can wire into their own products. The Macau stop matters institutionally — it places the launch inside the bilateral Portugal-Macau science and education calendar, the same channel that has been used to onboard Brazilian research groups during the closed beta.

For users in Portugal, the practical question is what Amália plugs into first. Government sources have repeatedly flagged the Administração Pública front of house — the e-Balcão case-management overlays on the Portal das Finanças, the chat layer that AIMA tested at the contact-centre pilot in late 2025, and a research-translation pipeline at the Direcção-Geral de Estatísticas da Educação e Ciência (Directorate-General for Education and Science Statistics, DGEEC). None of those routes is contractually disclosed in the launch package, but the open-licence frame is intended to let line ministries pick the model up without procurement re-negotiation.

The €5 million envelope is small versus comparable national-LLM programmes in France or Germany, where state-backed efforts have run into the low hundreds of millions. The Portuguese pitch from the FCT side has been about specialisation, not scale: train tightly on European-Portuguese corpora — legal, academic and administrative — and release with permissive terms so derivative tunings can be done downstream. That trade-off will be tested as soon as the weights are published. Early reviewers from the consortium have described the beta as “very positive” on Portuguese grammatical fidelity, but the public benchmark window will be short.

Three things to watch over the next fortnight. First, the licence text itself — Apache-2.0 versus a more restrictive research clause would materially change Amália’s downstream economics. Second, whether the launch ships an instruction-tuned variant or only the base model — the latter would push the work back to integrators. Third, the Brazilian uptake signal: Alexandre’s repeated “Lusophone” framing is also a soft-power claim, and a São Paulo or Rio research partner taking the weights in week one would validate the agenda far more than a Lisbon-only roll-out.

Sources: ECO, Público, Observador (15-16 June 2026).