Portugal open-sources Amália, its first national AI model for European Portuguese

A nation of just over ten million people now possesses a large language model that understands its own distinct dialect. Portugal has officially launched Amália, the country’s first national AI model purpose-built for European Portuguese, and it has done so with a level of government deliberation rarely seen in the field. The model, along with its underlying training data and the code used to build it, has been released under an open-source license.

Named after the legendary fado singer Amália Rodrigues, the model represents a significant step in linguistic preservation and technological sovereignty. Unlike many mainstream AI systems that are trained predominantly on Brazilian Portuguese, Amália has been refined specifically on the vocabulary, grammar, and cultural nuances of the European variant. This distinction matters for everything from formal government communications to everyday conversational tone.

The project was spearheaded by Portugal’s National Center for Artificial Intelligence in collaboration with several universities and tech partners. The team curated a dataset comprising millions of documents, including legal texts, literature, news articles, and web content, all sourced from Portugal. Every piece of data was carefully vetted for quality and relevance, ensuring the model does not simply regurgitate internet noise.

Because the model is open-source, any developer, researcher, or business can download it, inspect its architecture, and fine-tune it for specific applications. This transparency is a deliberate counterpoint to the closed, proprietary models offered by major tech companies. The Portuguese government hopes that Amália will spur local innovation, allowing startups and public institutions to build AI-powered tools without relying on foreign infrastructure or paying licensing fees.

Initial benchmarks show that Amália outperforms several general-purpose models on tasks involving European Portuguese, including reading comprehension, grammar correction, and cultural reference recognition. However, the model is not without limitations. Its training corpus is smaller than those used for English or Spanish models, and it may struggle with highly technical or niche vocabulary. The team behind Amália has acknowledged these gaps and plans to release updated versions as more data becomes available.

The release has been met with enthusiasm from linguists and technologists who argue that smaller languages are often overlooked in the AI boom. By open-sourcing the model, Portugal is not only serving its own citizens but also providing a blueprint for other nations with under-represented languages. The move aligns with a growing global push for sovereign AI systems that reflect local values and linguistic realities.

For now, Amália is available for download on GitHub and through several cloud platforms. The Portuguese government is also funding a series of workshops to help developers and educators learn how to use the model effectively. Whether Amália will achieve widespread adoption remains to be seen, but its existence alone marks a milestone: a small country refusing to let its language be an afterthought in the age of artificial intelligence.