Portugal has taken a significant step in the world of artificial intelligence by open-sourcing Amália, its first national large language model (LLM) built specifically for European Portuguese. The model, its training data, and source code are all freely available for governments, universities, and companies to use and build upon. This move positions Portugal as a frontrunner in the race for linguistic and technological sovereignty within the European Union.
What is Amália?
Amália is an acronym that stands for Automatic Multimodal Language Assistant with Artificial Intelligence. The name deliberately invokes Amália Rodrigues, the iconic fado singer whose voice is deeply intertwined with Portuguese cultural identity. The model is built on EuroLLM-9B, a European foundation model developed as part of the OpenEuroLLM initiative. A team of over 60 researchers and students expanded this base with European Portuguese datasets, a larger context window, strong safety and evaluation systems, and the ability to handle images alongside text.
Unlike commercial chatbots such as ChatGPT, Amália is not designed as a consumer-facing app. Instead, it is intended to serve as an underlying layer that other software can call upon. Planned applications include an AI teaching assistant, a virtual guide for Portuguese museums and monuments, a digital assistant for citizen services, and decision-support tools for the Portuguese Navy. This distinction explains why the Portuguese state is giving the model away rather than charging for access—it is a public infrastructure investment.
Funding and Development
The project has drawn an initial €5.5 million through Portugal’s Recovery and Resilience Plan. The funding flows 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 presented at the PROPOR conference in Brazil. Funding has been secured through the end of 2027, signaling a long-term commitment rather than a one-off launch.
The Importance of European Portuguese
One of the strongest selling points of Amália is its focus on European Portuguese. While Brazilian Portuguese is spoken by over 200 million people, the European variant has distinct differences in grammar, vocabulary, and cultural references. Major commercial AI models, trained predominantly on Brazilian Portuguese data, tend to flatten these differences. A model that accurately captures the nuances of European Portuguese is invaluable for public services that need to communicate with citizens in their own register. For example, a citizen interacting with a government chatbot about tax forms or healthcare expects precise and culturally appropriate language, not a generic approximation.
This linguistic specificity is not just about accuracy—it is about identity. Language is a cornerstone of national culture, and many Portuguese speakers feel that their version of the language is underrepresented in global AI systems. Amália addresses this gap directly, offering a tool that understands local idioms, expressions, and even regional dialects.
Open Source and Sovereign AI
By open-sourcing the model, Portugal is embracing a philosophy of transparency and independence. The weights, datasets, and code are published under an open license, allowing anyone to inspect how the model was trained, adapt it, and run it on their own hardware. This choice is both ideological and practical. A model intended for use in citizen services and naval decision-support must be auditable, not merely trusted. Open publication is the surest way to guarantee that option.
The release lands squarely inside Europe’s wider unease about depending on American and Chinese AI systems for something as foundational as language. The reliance on a handful of large tech companies for AI infrastructure has raised concerns about data privacy, sovereignty, and cultural homogenization. Amália is part of a broader European push to develop homegrown AI capabilities, alongside projects like OpenEuroLLM and infrastructure investments such as Nscale’s €695 million data-centre push in Portugal and Microsoft’s cloud expansion. However, some critics argue that renting GPUs by the hour can create an illusion of sovereignty rather than the real thing. Amália’s open nature is intended to address this by giving users full control over the model and its deployment.
Challenges and Adoption
While the technical achievement is notable, the harder question is adoption. Publishing a model openly is one thing; getting universities, companies, and government departments to actually build on it is another. Many sovereign AI projects have quietly run out of steam at this second phase. Portugal has funded Amália through 2027 and named the institutions meant to carry it forward, but sustained adoption will require active community building, documentation, and support. The next two years will show whether Amália becomes real infrastructure for Portugal or remains a well-documented research project with a evocative name.
Another challenge is competition from larger, more established commercial models. Even if Amália outperforms them in European Portuguese, users may still prefer the broader capabilities of ChatGPT, Claude, or Gemini for tasks not limited to Portuguese. To succeed, the ecosystem around Amália will need to offer compelling integrations, user-friendly tools, and clear benefits for developers. The open-source community can play a key role here, building plugins and applications that leverage the model’s strengths.
Broader Context: Europe’s AI Ambitions
Portugal’s move is emblematic of a larger trend in Europe: the quest for digital sovereignty. The European Union has invested heavily in AI research, data infrastructure, and regulatory frameworks like the AI Act. However, the continent still lags behind the United States and China in terms of commercial AI adoption and talent acquisition. Initiatives like Amália aim to close that gap by focusing on specific, culturally relevant use cases rather than trying to compete directly with Silicon Valley giants. The success of such projects could pave the way for other European nations to develop their own national AI models for languages like Dutch, Swedish, or Catalan, creating a mosaic of localized AI systems that collectively strengthen Europe’s technological independence.
The open-source nature of Amália also aligns with the EU’s values of transparency, accountability, and collaboration. By making the model publicly available, Portugal invites scrutiny and contribution from the global research community, which can help identify biases, improve performance, and expand applicability. This approach contrasts sharply with the closed, proprietary models that dominate the current AI landscape.
Technical Details and Capabilities
Amália is built on EuroLLM-9B, which itself is a large language model with 9 billion parameters trained on multilingual European data. The Portuguese-specific fine-tuning involved curating high-quality datasets from Portuguese literature, government documents, news articles, and social media. The expanded context window allows the model to handle longer documents, such as legal texts or scientific papers, without losing coherence. Safety systems were also enhanced to prevent harmful outputs, including those that might propagate stereotypes or misinformation about Portuguese culture.
Multimodality is another key feature: Amália can process both text and images, enabling applications like analyzing museum artworks or interpreting diagrams in educational materials. This capability is achieved through a vision encoder that maps images into the same embedding space as text, allowing the model to perform tasks like visual question answering or image captioning with a Portuguese twist.
The model’s training infrastructure used a cluster of GPUs provided by Portuguese academic institutions, with computational resources supplemented by cloud services from European providers. This hybrid approach kept costs manageable while ensuring that the training process remained under national control. The entire pipeline, from data collection to model release, was documented in a transparent manner, setting a precedent for future state-funded AI projects.
Cultural Significance
Naming the model after Amália Rodrigues is a deliberate cultural statement. Rodrigues is arguably Portugal’s most famous cultural export, known for her powerful interpretations of fado, a genre that embodies the country’s soulful and melancholic spirit. By associating the AI with her legacy, the Portuguese government signals that Amália is not just a technical tool but a cultural artifact that should reflect and preserve Portuguese identity. This move also helps build public trust and enthusiasm for AI, framing it as a homegrown innovation rather than a foreign imposition.
The choice of a female name for the AI is also notable. In a field often dominated by masculine metaphors and terminology, Amália brings a touch of the feminine, echoing the historical role of women in Portuguese culture as keepers of language and tradition. Whether this will have a concrete impact on user perception remains to be seen, but it is a thoughtful touch in a project that clearly values symbolism alongside substance.
Portugal has positioned Amália as a model for other nations seeking to reclaim linguistic and technological autonomy. The upcoming years will determine whether this open investment leads to widespread adoption or remains a proud but underutilized prototype. For now, Amália stands as a testament to what a small country can achieve with focused funding and collaborative academic effort, challenging the notion that only large corporations or nations can lead in AI.