Portugal's release of Amalia is a working template for how a smaller country can ship a sovereign, open language model on a research-project budget rather than a hyperscaler one, a pattern worth noting for teams evaluating national or regional AI deployments. The Portuguese government unveiled Amalia on July 1 as the first open large language model built for European Portuguese, developed by a consortium of universities led by NOVA University Lisbon with more than sixty researchers, and funded through the country's Recovery and Resilience Plan at a cost rising to 7 million euros by 2027. Rather than training from scratch, the team adapted EuroLLM-9B and the earlier GlorIA model, extending pretraining and expanding context to 32,000 tokens, then released the roughly 9-billion-parameter text model, a vision model, and a speech-recognition component under the Apache 2.0 license on Hugging Face. The approach mirrors earlier national efforts including Spain's ALIA and Germany's Teuken-7B.
The secret AI that could help explain this wild World Cup