📊 Full opportunity report: AMÁLIA · The Three Hard Questions. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Portugal’s AMÁLIA, a €5.5 million European Portuguese LLM, is now operational. However, critical questions about its openness, native-language data, and optimization goals remain unresolved, highlighting broader issues in European AI sovereignty.
Portugal’s €5.5 million AMÁLIA large language model is now operational, with a base version released and performance benchmarks surpassing many prior models on Portuguese tasks. However, fundamental questions about its openness, data sufficiency, and strategic goals remain unanswered, raising concerns about the broader European sovereign-LLM movement.
The AMÁLIA project involves approximately 60 researchers from Portugal’s leading academic institutions, including NOVA, IST, and IT. It was announced in December 2024, with the model completed by September 2025 and publicly launched in October 2025. The model is accessible to 450,000 academic users via the FCT’s IAedu platform and is trained on a mixture of multilingual and Portuguese-specific data, with knowledge up to the end of 2023.
Technically, AMÁLIA is a continuation of the EuroLLM model, not trained from scratch, which distinguishes it from other European projects like Italy’s Minerva. Its training involved 107 billion tokens, with around 5.8 billion tokens from Portuguese web archives, representing roughly 5.5% of the extended pre-training data. The model outperforms previous open models on Portuguese benchmarks and exceeds Qwen 3-8B on most tests, though it still trails on some specific benchmarks such as ALBA.
Despite these achievements, questions about the model’s openness, native-language data sufficiency, and strategic optimization remain largely unaddressed, sparking debate about the true state of European AI sovereignty and the structural challenges faced by national LLM initiatives.
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.
AI Systems Performance Engineering: Optimizing Model Training and Inference Workloads with GPUs, CUDA, and PyTorch
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.
Portuguese for Beginners: Practical Learning with SynapseLingo (Learn Portuguese)
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.
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.
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 AI Sovereignty and Policy
The development and deployment of AMÁLIA underscore the broader challenges faced by European countries in establishing sovereign AI capabilities. While Portugal’s investment demonstrates commitment, the unresolved questions about openness, native data, and strategic goals reveal systemic issues that could affect future national and regional AI initiatives. Addressing these questions is critical for ensuring transparency, competitiveness, and alignment with European data governance standards, influencing the continent’s position in global AI development.
Structural Challenges in European Sovereign LLM Projects
European countries have launched multiple large language model projects—such as Italy’s Minerva, Germany’s Aleph Alpha, and France’s Mistral—each grappling with similar core questions. These projects often operate with limited native-language data, and their openness varies, raising concerns about transparency and strategic coherence. The European sovereign-LLM movement is characterized by a pattern of ambitious investments but lacks a unified framework to address fundamental questions about data, openness, and goal-setting, which are now surfacing in Portugal’s AMÁLIA case.
Historically, most models are built on multilingual foundations, with native-language adaptation as an afterthought. The debate over whether to train from scratch or continue from existing multilingual models remains unresolved, affecting the models’ performance and strategic autonomy. Portugal’s approach, continuing from EuroLLM, exemplifies this pattern and prompts reflection on the broader European strategy.
“The AMÁLIA case exposes three fundamental questions that European sovereign-LLMs must answer publicly—questions about openness, native data, and strategic goals—that are currently being overlooked.”
— Thorsten Meyer
Unanswered Questions About AMÁLIA’s Openness and Goals
It remains unclear how open AMÁLIA will be in practice, especially regarding access to its training data and model weights. The extent to which native Portuguese data has been prioritized or curated is also not fully disclosed. Additionally, the strategic objectives—whether the model is meant to serve as a national AI asset, a research tool, or a commercial product—are still ambiguous. The final version, expected in June 2026, may address some of these gaps, but current details are limited.
Next Milestones and Transparency Efforts for AMÁLIA
In the coming months, the AMÁLIA team is expected to release more detailed documentation and possibly open-source components, which could clarify questions about data and openness. The final version scheduled for June 2026 will be a key milestone to assess whether the project has addressed these core questions. Additionally, broader European discussions on sovereignty and model transparency are likely to influence Portugal’s ongoing development and policy decisions regarding AMÁLIA and similar models.
Key Questions
Will AMÁLIA be fully open-source?
It is not yet clear whether the final version will be fully open-source. The current disclosures do not specify access to training data or model weights, and transparency remains a key concern.
How much native Portuguese data was used in training AMÁLIA?
Approximately 5.8 billion tokens from Portuguese web archives were used, representing about 5.5% of the total extended pre-training data. The significance of this proportion for model performance is still debated.
What are the strategic goals behind AMÁLIA?
The Portuguese government has not publicly clarified whether the model aims to serve as a national AI infrastructure, a research platform, or a commercial product, leaving this question open for interpretation.
Will Portugal continue to develop native-language models beyond AMÁLIA?
Future plans are not yet publicly detailed, but the ongoing development of AMÁLIA and similar initiatives suggest a strategic interest in maintaining sovereignty in AI capabilities.
Source: ThorstenMeyerAI.com
