Amália

Amália is an open-source large language model built specifically for European Portuguese, released on July 1, 2026 by a government-backed Portuguese university consortium. The weights, training dataset and code are published on Hugging Face under an open license, so anyone can download, fine-tune, self-host or audit the model rather than reach it only through a paid API.

The pitch is sovereignty, not benchmark leadership. Most frontier models treat Portuguese as an afterthought and skew toward Brazilian usage; Amália is trained on European Portuguese data so its grammar, vocabulary and cultural references match how people actually write in Portugal. It's part of a wider European push — alongside efforts in France, Spain and the Nordics — to keep public-sector and language-specific AI on infrastructure that governments and researchers can inspect and control, instead of depending entirely on US labs.

Because it ships as open weights rather than a polished consumer app, the audience is developers, public institutions, and researchers who want a Portuguese-native base model to build on. It's closer in spirit to Meta's Llama or Mistral's open releases than to ChatGPT: the value is the checkpoint and the transparent dataset, not a turnkey chat product. If you need a plug-and-play assistant, a hosted model will be smoother; if you need a Portuguese-language model you can run and modify yourself, this is a rare option.