Lisbon, July 1, 2026 (Lusa) - The chief executive (CEO) of the Centre for Responsible AI emphasised in an interview with Lusa that Amália is a specialised system designed for specific applications rather than a conversational system like ChatGPT, and that the organisation will make it available to anyone working in the field of artificial intelligence.
“Amália will be completely open in the sense of what is known as an open and responsible model. What does this mean? It means that anyone can integrate Amália [provided] they work in this field of artificial intelligence,” said Paulo Dimas, CEO of the Centre for Responsible AI.
“The Amália team will focus on providing an open, responsible foundation model rather than a chat interface like ChatGPT, because its purpose lies in enabling broader AI innovation. It’s important to realise that this isn’t ChatGPT, and it’s very important to repeat this, because people usually have in their minds those initial moments of wonder that we all felt when using ChatGPT, and now we think, ‘and now we’re going to have this in Portuguese, made in Portugal’,” emphasises Paulo Dimas.
Amália, a large language model (LLM), will be presented this afternoon at the Técnico Innovation Centre in Lisbon.
“It is a specialised system designed for specific tasks. It is an artificial intelligence tool that ensures “three levels of sovereignty [language, culture and data]” and which will be freely available; this means that anyone can download it, anyone who wishes to apply this artificial intelligence,” he stressed.
Paulo Dimas points out that “it is a model that is open in three ways”.
Firstly, “it is open to being extended. Let’s imagine we now want to take Amália and apply it, say, to a legal field; we can build on the Amália framework and create a model for that” or, if we wanted to “create a Eusébio: anyone could take the approach used to build Amália, where all the knowledge is shared, and apply it to Eusébio”, he explained.
Or even someone who wants to look at the data “and see what data was used and use that data for various applications can do so too”, he said.
As for the ordinary citizen, they will benefit indirectly when “they’re interacting with the gov.pt assistant, which is the Public Administration portal, the portal developed by ARTE – the Agency for State Technological Reform”, he explained.
For example, “if you’re a teacher, and you’re using the system being developed at the University of Minho by Paulo Novais’s team, you’ll be able to benefit from Amália helping you to plan your lessons, write summaries and ensure the quality of the Portuguese text it generates is good”.
Even a start-up can benefit from Amália.
“Several start-ups have already approached us, aware of our connection to the consortium from the outset,” to find out how they can start using Amália in their products because customers want to ensure that the Portuguese remains distinct from Brazilian Portuguese, he noted.
Paulo Dimas emphasised that “the model is still in its early stages in terms of certain functions that have been added to other models, but it is a foundation”.
This model “is a relatively small model, with 9 billion parameters, but a model with 22 billion parameters is already on the roadmap, which the consortium will subsequently announce”, he said.
“A very important point is that Amália exists thanks to another LLM, namely EuroLLM”, which the team built from scratch for the languages of the European Union.
“I think it’s incredible that we’re experiencing this moment now and have here an example of a project that wasn’t developed by the Centre [for Responsible AI], it’s extremely important that this is made very clear, it was merely, shall we say, that the initial spark was ignited at the Centre, given the research in this area, which in 18 months met the objectives exactly as planned,” he noted.
He acknowledged that “Amália has had many detractors” and will continue to have them, but as a professional passionate about the project, he said “it should be a national flagship for artificial intelligence”.
As for whether Amália will sing, Paulo Dimas says: “Not directly, but through a complementary model developed by Paulo Romano’s team at Inesc-ID; she will also be able to speak, but she will likely leave singing to other applications.”
ALU/ADB // ADB.
Lusa
