📊 Full opportunity report: AMÁLIA · The Three Hard Questions. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Portugal’s national LLM, AMÁLIA, is now operational and outperforms several benchmarks, yet fundamental questions about its openness, native-language data, and objectives remain unresolved. These issues impact the broader European sovereign-LLM landscape.
Portugal’s €5.5 million AMÁLIA large language model is now operational, with the base version publicly available and outperforming several benchmarks in European Portuguese tasks. This development marks a significant step for Portugal’s AI ambitions but raises pressing questions about the model’s openness, native-language data, and strategic objectives, which remain largely unanswered.
AMÁLIA was developed through a consortium involving approximately 60 researchers from Portugal’s top research institutions, including NOVA, IST, and IT, and was announced in December 2024. The model, which handles text only in its current form, was completed on September 30, 2025, and is accessible via the FCT’s IAedu platform to 450,000 academic users. It is based on a continuation of the EuroLLM multilingual foundation, with a training process that included 107 billion tokens, of which only about 5.8 billion tokens originated from Portuguese sources.
According to the technical report by Vieira et al. (2026) and analysis by Duarte O.Carmo, AMÁLIA outperforms previous open models on European Portuguese benchmarks and surpasses Qwen 3-8B on most tests, although it still trails on certain specific benchmarks like ALBA. The final version is expected in June 2026, with ongoing development and refinement.
AMÁLIA
The three hard
questions.
Portugal spent €5.5M to build a European Portuguese LLM. The base version is operational, the benchmarks beat Qwen 3-8B on most pt-PT tasks. So why are the most important questions still unanswered?
Last month, Duarte O.Carmo published the sharpest public analysis of AMÁLIA — Portugal’s state-funded European Portuguese large language model. He prefaces his critique with the necessary diplomatic apparatus before doing what almost nobody else in the European-sovereign-LLM discourse has been willing to do publicly: asking hard questions about whether the work, as released, actually does what it set out to do. This piece is a structural extension of his analysis. The AMÁLIA case study exposes three hard questions every national LLM effort needs to answer publicly — and the broader European sovereign-LLM movement has been operating without explicit answers to any of them.
Three questions every national LLM effort needs to answer publicly.
Duarte O.Carmo’s framing maps cleanly onto the structural argument. Each question lands specifically in AMÁLIA — and the broader European sovereign-LLM movement has been operating without explicit answers to any of them.
The three questions form a structural feedback loop. Q3 (optimization target) determines Q2 (data volume needed) which conditions Q1 (openness sufficient for community contribution). The European sovereign-LLM movement collectively benefits from these questions becoming standard methodology disclosure, not exceptional critique.
Advanced Language Tool Kit: Teaching the Structure of the English Language
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
107 billion tokens. 5.8 billion clearly pt-PT.
The structurally tractable question with a structurally surprising answer. For a model whose entire stated purpose is European Portuguese prioritization, the native-language share of extended pre-training is 5.5%. The implications cascade into every other question.
European Portuguese NLP training data
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Olmo standard. AMÁLIA’s current state.
Allen Institute for AI’s Olmo project defines what “fully open” operationally requires. Olmo doesn’t lead frontier benchmarks. That’s not the point. The point is to be the structural reference for openness. AMÁLIA’s “fully open source” claim should track to the operational standard.
Evals for AI Engineers: Systematically Measuring and Improving AI Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Four strategic positions. AMÁLIA between two and three.
Approximately €100M+ in publicly disclosed European sovereign-LLM funding across the major initiatives. The structural question every project faces: what is the actual competitive position you’re staking? Four options — none mutually exclusive — but each requiring different commitments.
Open Source AI Models On Mobile: Deploying Lightweight LLMs On Android And iOS Devices
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Three standards. For AMÁLIA and the movement.
The structural critique generalizes beyond AMÁLIA. Italy, France, Germany, Switzerland, the OpenEuroLLM consortium, and every subsequent national project benefit from public discourse holding national LLM efforts to operational standards on openness, data accounting, and strategic positioning.
The European sovereign-AI agenda is a serious strategic project that deserves serious public discourse. O.Carmo’s analysis is what serious public discourse looks like. Appropriately diplomatic. Structurally rigorous. Willing to ask the hard questions in public when the public investment justifies it. More of this is needed — across every European sovereign-LLM project, not just AMÁLIA.
Implications for European Sovereign-Language Models
The development of AMÁLIA exemplifies Portugal’s strategic investment in sovereign-language AI, positioning it within a broader European effort to create models tailored to national languages. However, the project’s progress highlights critical issues that could influence policy and future model development across Europe, especially concerning transparency, data sourcing, and strategic priorities.
These questions matter because they determine how national models serve their linguistic communities, compete globally, and align with public interests. The answers will shape the future of European AI sovereignty and influence how governments allocate resources and set standards for open and responsible AI development.
European Sovereign-Language Model Initiatives and Challenges
Across Europe, multiple countries are pursuing their own LLM projects, including Italy’s Minerva, Germany’s Aleph Alpha, France’s Mistral, and initiatives like OpenEuroLLM, AI Sweden, and Apertus. These efforts are driven by a shared desire to foster national AI sovereignty amid concerns over reliance on U.S. and Chinese models. A common pattern emerges: models are often based on multilingual foundations, with native-language data sources varying significantly in size and quality.
Despite substantial investments, questions about how open these models truly are, how much native-language data is sufficient, and what objectives they should prioritize remain largely unaddressed publicly. The European discourse often treats individual model launches as isolated events, rather than parts of a broader structural pattern that raises systemic questions about transparency and strategy.
“The three questions—about openness, native data, and objectives—are fundamental to understanding the true nature of our sovereign LLM efforts.”
— Duarte O.Carmo
Unanswered Questions About AMÁLIA’s Strategic and Technical Foundations
It remains unclear how open AMÁLIA truly is beyond its technical documentation, especially regarding access to training data and model weights. The extent to which native Portuguese data influences its capabilities and strategic decisions about future development are still under discussion. Additionally, the final objectives of the project—whether it aims primarily for academic use, commercial deployment, or national policy—are not yet explicitly defined.
Furthermore, it is uncertain how these questions will be addressed before the final release in June 2026, and whether the project will adapt based on ongoing evaluations and external critiques.
Next Milestones and Policy Implications for AMÁLIA
The immediate next step is the release of the final version of AMÁLIA in June 2026, which will likely include further refinements and possibly more transparent disclosures. Over the next 12-24 months, Portugal and other European nations will assess whether their models meet strategic goals, especially regarding openness, native-language data sufficiency, and alignment with public interests.
Public and academic scrutiny, along with policy debates, are expected to intensify, potentially influencing future funding, transparency standards, and collaborative frameworks for sovereign-language AI projects across Europe.
Key Questions
What makes AMÁLIA different from other European language models?
AMÁLIA is based on a continuation of a multilingual foundation, with a focus on Portuguese, and is developed through a public-private consortium involving Portugal’s top research institutions. Its development emphasizes native-language data and national strategic interests.
Why are questions about openness and native data important?
These questions determine how transparent the model’s development process is, how well it represents the native language, and whether it aligns with national or public interests. Openness affects trust, usability, and future collaboration.
What are the main concerns critics have about AMÁLIA?
Critics question whether the model is truly open, whether enough native Portuguese data has been used, and what the strategic goals are. They worry about transparency and whether the project prioritizes academic, commercial, or policy objectives.
How does AMÁLIA compare to models like Qwen or Minerva?
AMÁLIA outperforms some open models on Portuguese benchmarks and beats Qwen 3-8B on most tests, but still trails on certain benchmarks like ALBA. Unlike Minerva, which was trained from scratch, AMÁLIA is a continuation of a multilingual pre-training, which influences its data and strategy choices.
Source: ThorstenMeyerAI.com
