Portugal launches AMÁLIA, a free open-source AI language model built for European Portuguese, backed by €5.5 million and a 18-month development plan

Portugal's government announced AMÁLIA (Assistente Multimodal Automático de Linguagem com Inteligência Artificial), the country's first large language model (AI trained on massive text data to understand and generate language) optimized for Portuguese. Development began November 11, 2024, with a €5.5 million investment and will deliver three versions: a beta version by Q1 2025, a base version by Q3 2025, and a multimodal version (handling text, images, and video) by Q2 2026.

Unlike most commercial AI models that perform poorly outside English, AMÁLIA distinguishes between Portuguese language variants, understands Portuguese culture and history, and keeps sensitive data stored locally—meaning government agencies and hospitals won't need to share private patient or citizen records with foreign tech companies to use the system.

Portuguese businesses, researchers, schools, and public services can use AMÁLIA for free, in open-source form. This gives companies in education, healthcare, and public administration a tool they can customize for their own needs without licensing fees or vendor lock-in—similar to how Spain built ALIA for Spanish, Catalan, Galician, and Basque.

The first beta version arrives in Q1 2025 as a text-only chatbot trained on data from Arquivo.pt (Portugal's web archive). All code and training data will be publicly released, letting Portuguese developers build on it rather than relying solely on OpenAI's or Google's models.