Portugal has released Amália, its inaugural national AI model, as an open-source initiative focused on European Portuguese.

      A nation with a population of just over 10 million has now introduced a large language model that represents its unique version of the language. Portugal has unveiled Amália, its first national AI model specifically designed for European Portuguese, and has done so in a highly methodical manner typical of governmental initiatives.

      The model, along with its training data and source code, is entirely open and available at no cost for governments, universities, and corporations to utilize and build upon. The name is significant; Amália stands for Automatic Multimodal Language Assistant with Artificial Intelligence, and it also pays homage to Amália Rodrigues, the fado singer whose voice is intertwined with Portuguese identity.

      The model is based on EuroLLM-9B, a European foundation model, which has been enhanced by a team of over 60 researchers and students with European Portuguese datasets, an expanded context window, improved safety and evaluation mechanisms, and the capability to process images alongside text.

      What Amália is not is a competitor to ChatGPT as most people would typically engage with one. It will not be launched as a consumer chat application. Rather, it is intended to operate as an underlying framework that other software can utilize.

      Projected applications include an AI teaching assistant, a virtual guide for Portuguese museums and monuments, a digital assistant for governmental services, and decision-support tools for the Portuguese Navy. This distinction clarifies why the state opts to offer the model for free instead of charging for access.

      The project has initially received €5.5 million financing from Portugal’s Recovery and Resilience Plan, with funds allocated to NOVA University Lisbon, Instituto Superior Técnico, and the universities of Porto, Minho, and Coimbra, coordinated with the Foundation for Science and Technology. A test version was completed in September 2025 and showcased at the PROPOR conference in Brazil. Funding has already been assured until the end of 2027, indicating a commitment to maintaining the project rather than merely launching it.

      In this context, "open" indeed means open. Unlike large commercial systems that operate as closed boxes accessed only via an interface and payment, Amália is provided with its weights, datasets, and code released under an open license, allowing anyone to examine how it was trained, modify it, and run it on their own infrastructure. This decision is motivated by both ideological considerations and practical needs.

      A model that a government plans to integrate into citizen services and naval decision-support requires the ability to audit it, not simply trust it, and open publication is the most reliable method to ensure that option remains available. The release coincides with Europe's broader concerns about reliance on American and Chinese systems for something as fundamental as language processing.

      It follows the OpenEuroLLM alliance, a cross-border initiative to train open models on Europe's own languages, alongside a series of infrastructure investments, including Nscale’s €695 million data center project in Portugal in partnership with Microsoft. However, the question of whether this truly leads to genuine independence remains debated.

      Renting GPUs by the hour, as argued by TNW, may create an illusion of sovereignty without the true essence of it.

      Amália’s most significant advantage is its specificity. European Portuguese differs from Brazilian Portuguese, and the major commercial models, which are predominantly trained on Brazilian Portuguese, often fail to recognize these distinctions. A system that accurately captures grammar, idioms, and cultural references proves to be useful in a way that a larger, less precise model does not, which is particularly crucial for public services that need to communicate with citizens in their own language rather than a mere approximation.

      The more challenging issue lies in adoption. While releasing a model openly is one thing, encouraging universities, corporations, and government departments to actively build on it is another, and that second phase is where many sovereign AI initiatives struggle to progress.

      Portugal has committed funding to Amália through 2027 and designated the institutions responsible for advancing it. The next two years will determine whether it evolves into a genuine infrastructure or remains just a well-documented research initiative with an appealing name.