📊 Full opportunity report: AMÁLIA · The Three Hard Questions. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Portugal launched AMÁLIA, a European Portuguese large language model, with promising benchmarks. However, critical questions about its openness, native data sufficiency, and optimization goals remain unresolved, raising concerns about the broader European sovereign-LLM movement.
Portugal’s €5.5 million investment in the large language model AMÁLIA has resulted in the release of its base version, which outperforms most open models on European Portuguese benchmarks and surpasses Qwen 3-8B on several tasks. However, fundamental questions about the model’s openness, native data adequacy, and optimization priorities remain unanswered, highlighting broader issues within the European sovereign-LLM initiative.
AMÁLIA is a consortium project involving approximately 60 researchers from Portugal’s top research institutions, including NOVA, IST, IT, and FCT. The model was completed in September 2025 and is currently accessible to 450,000 academic users via the FCT’s IAedu platform. It is based on a continuation of the pre-training phase of EuroLLM, rather than training from scratch, with the training pipeline including 107 billion tokens, of which only about 5.8 billion are Portuguese, primarily from the national web archive Arquivo.pt.
Benchmark results show AMÁLIA surpasses previous open models on European Portuguese tasks and beats Qwen 3-8B on most benchmarks, though it still trails Qwen on the primary European Portuguese benchmark, ALBA. The final version is scheduled for release in June 2026, and the team has indicated ongoing development that may address current gaps.
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.
Romans in a New World: Classical Models in Sixteenth-Century Spanish America (History, Languages, And Cultures Of The Spanish And Portuguese Worlds)
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.
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.
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.
Training Data for Machine Learning: Human Supervision from Annotation to Data Science
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.
Agentes de IA na Prática: Como Criar Agentes de Inteligência Artificial Autônomos (Portuguese Edition)
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.
Structural Questions in European Sovereign LLMs
This development underscores the broader challenge faced by European countries in developing sovereign language models. Despite significant investments, critical questions about how open these models truly are, whether native-language data is sufficient, and what their primary optimization goals should be remain unresolved. These issues impact national policy decisions and the future of European AI sovereignty, influencing how models are built, evaluated, and deployed across the continent.
European Sovereign LLM Initiatives and Strategic Challenges
Over the past year, multiple European nations and consortia, including Italy’s Minerva, Germany’s Aleph Alpha, France’s Mistral, and others, have launched or announced large language models with public funding. These efforts are part of a broader movement to develop independent, sovereign AI capabilities amid concerns over reliance on US and Chinese models. However, most initiatives face similar structural questions about openness, native data use, and goal-setting, which remain under-discussed publicly.
Portugal’s AMÁLIA is notable for its public investment and the involvement of top research institutions, making it a key case study for understanding the structural challenges faced by the European sovereign-LLM movement. The ongoing debate centers on whether these models can truly be open, how much native data is enough, and what objectives they should prioritize—issues that are critical for policy and strategic planning.
“The AMÁLIA project raises fundamental questions about openness, native data sufficiency, and the core objectives of European language models.”
— Duarte O.Carmo
Unresolved Questions About Openness, Data, and Goals
It is still unclear how open AMÁLIA truly is, given the reliance on a continuation of a multilingual foundation and limited native Portuguese data. The sufficiency of the 5.8 billion tokens from Arquivo.pt for native-language mastery remains unconfirmed. Additionally, the primary objectives guiding the model’s development—whether for academic, commercial, or strategic national purposes—are not explicitly defined or publicly discussed.
While the final version is scheduled for June 2026, it is not yet clear whether these gaps will be addressed or if new issues will emerge as development progresses.
Next Steps for Portugal’s AMÁLIA and European Sovereign LLMs
The upcoming months will see further development and testing of AMÁLIA, with the final version expected in June 2026. Researchers and policymakers will likely scrutinize the model’s openness, native data integration, and objectives more closely. Broader European initiatives may also reassess their strategies in light of these structural questions, potentially leading to new standards or collaborative frameworks for sovereign LLM development.
Public discussions and transparency around these core issues are expected to increase, shaping the future landscape of European AI sovereignty efforts.
Key Questions
What makes AMÁLIA different from other European language models?
AMÁLIA is based on a continuation of a multilingual foundation rather than training from scratch, with a focus on Portuguese and a significant investment from the Portuguese government. Its benchmarks show competitive performance, but questions about openness and native data use remain.
Why are questions about openness and native data important?
Openness determines how accessible and transparent the model is, affecting trust and collaboration. Native data relevance impacts the model’s accuracy and cultural fidelity, which are crucial for national sovereignty and language preservation.
What are the main risks of these unresolved questions?
Uncertainty about openness and data sufficiency could limit the model’s adoption, hinder transparency, and affect strategic national interests. It may also impact the credibility of European efforts to develop independent AI capabilities.
Will the final version of AMÁLIA address these questions?
It is not yet clear whether the final version will resolve these issues. Ongoing development may include measures to improve openness and native data integration, but details are still emerging.
What does this mean for Europe’s AI sovereignty?
These questions highlight the need for clearer strategic frameworks and transparency in European AI projects. Addressing them is essential for ensuring that sovereignty efforts are effective and trustworthy.
Source: ThorstenMeyerAI.com
