model release

Portugal releases Amália, open-source 9B parameter AI model trained on European Portuguese

TL;DR

Portugal has released Amália, its first national AI model trained specifically for European Portuguese. Built on EuroLLM-9B with 9 billion parameters, the model is fully open-source with weights, datasets, and code published under an open license. The government has committed €5.5m in initial funding through 2027.

2 min read
0

Portugal releases Amália, open-source 9B parameter AI model trained on European Portuguese

Portugal has released Amália, its first national AI model trained specifically for European Portuguese, with the model's weights, training data, and source code all published under an open license. The 9 billion parameter model is built on EuroLLM-9B, a European foundation model, and represents a €5.5m initial investment through Portugal's Recovery and Resilience Plan.

Technical specifications

Amália—Automatic Multimodal Language Assistant with Artificial Intelligence—is based on the EuroLLM-9B foundation model. A team of more than 60 researchers and students expanded it with European Portuguese datasets, added a larger context window than the base model, strengthened safety and evaluation systems, and built in multimodal capabilities to handle both images and text.

The test version was completed in September 2025 and presented at the PROPOR conference in Brazil. Funding is secured through the end of 2027.

Deployment model: Infrastructure, not consumer product

Amália is not designed as a consumer-facing chatbot. Instead, it functions as foundational infrastructure that other software can call on. 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.

The model is distributed through NOVA University Lisbon, Instituto Superior Técnico, and the universities of Porto, Minho, and Coimbra, coordinated with the Foundation for Science and Technology.

European Portuguese vs Brazilian Portuguese

European Portuguese differs significantly from Brazilian Portuguese in grammar, idiom, and cultural references. Large commercial models, trained predominantly on Brazilian Portuguese, tend to flatten these distinctions. Amália addresses this gap for Portugal's 10 million citizens and public services that require accurate cultural and linguistic register.

The release follows the OpenEuroLLM alliance and broader European efforts to reduce dependence on American and Chinese AI systems for foundational language tasks. It coincides with infrastructure investments including Nscale's €695m data-centre expansion in Portugal with Microsoft.

Open license details

Unlike commercial models accessed through APIs and usage fees, Amália ships with its weights, datasets, and code published under an open license. Governments, universities, and companies can inspect training data, adapt the model, and run it on their own hardware without recurring costs or dependencies on external providers.

This approach enables the auditing and verification that Portugal's government considers necessary for systems integrated into citizen services and military applications.

What this means

Portugal's release of Amália represents a concrete implementation of digital sovereignty—not just stated policy but functioning infrastructure with multi-year funding commitments. The model's value lies in its specificity: accurate European Portuguese for a defined user base, not a scaled-down competitor to GPT-4 or Claude. Success depends on actual adoption by universities, companies, and government departments over the next two years. Publishing the model is the easy part; integration into production systems is where most national AI projects stall. Portugal has structured funding and institutional coordination through 2027, which provides a realistic timeline to determine whether Amália becomes operational infrastructure or remains a well-documented research artifact.

Related Articles

model release

Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context

Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.

model release

Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license

Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.

model release

Tencent releases Hy4 flash: 220B MoE, 12B active parameters, 1M-token context

Tencent has released Hy4 flash, a multimodal Mixture-of-Experts model with 220B total parameters, 12B active per token, and a 1,048,576-token context window. Tencent calls it its most cost-effective Hy model to date and claims a per-task cost of 1/20 that of Hy4 on its internal WorkIndex benchmark.

model release

OpenAI releases GPT-Live-1.1 in the API, with standby mode billing idle WebRTC time at $0.02 per minute

OpenAI has released GPT-Live-1.1 in the API. The full-duplex voice model improves multilingual accents and transcription, and a new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05.

Comments

Loading...