📊 Full opportunity report: AMÁLIA · The Three Hard Questions. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Portugal’s AMÁLIA, a €5.5M European Portuguese LLM, is operational but faces three key questions about its openness, native-language data sufficiency, and optimization goals. These issues highlight broader challenges in European sovereign AI initiatives.
Portugal’s €5.5 million investment in the large language model AMÁLIA has resulted in an operational base version, but key questions about its openness, native-language data, and optimization priorities remain unresolved, raising broader concerns about European AI sovereignty efforts.
AMÁLIA is a consortium project involving approximately 60 researchers from Portugal’s top research institutions, launched by the government in December 2024. The model, built as a continuation of the EuroLLM multilingual foundation, was completed on September 30, 2025, and is currently accessible to 450,000 academic users via the FCT IAedu platform. It holds knowledge up to the end of 2023 and is expected to have a final version by June 2026.
The technical approach involves extending an existing multilingual model rather than training from scratch, contrasting with Italy’s Minerva, which was trained from the ground up on Italian and English data. AMÁLIA’s training included approximately 107 billion tokens, with only about 5.8 billion tokens from Portuguese web archives, representing roughly 5.5% of the extended pre-training. Despite outperforming previous open models on Portuguese benchmarks and beating Qwen 3-8B on most tests, it still trails Qwen on some key benchmarks like ALBA, the team’s primary European Portuguese benchmark.
However, questions remain about the model’s openness, the sufficiency of native-language data, and the primary goals of the project, as highlighted by recent critical analysis from Duarte O. Carmo. These questions are not accusations but fundamental to evaluating the project’s strategic success and European AI sovereignty.
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.
Jetson AGX Orin 64GB Developer Kit 275 Tops, with 1TB SSD,8MP USB Camera, AI Embedded Development Provides AI Large Models
AGX Orin 64GB Development Kit makes it easy to get started with AGX Orin. Its compact size, rich…
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.
MAYAPHILOS 224 Words Brazil Portuguese and English Talking Flash Cards for Toddlers, Autism Sensory Toys, Portuguese Language Learning Educational Montessori Speech Therapy Toys Gifts for Kids
【English and Portuguese Learning】The talking flash cards contain 510 sight words with 31 themes, including letters, numbers, animals,…
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.
AI model openness evaluation tools
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.
Learn to Speak Italian Language PC/Mac Software – Intermediate level for all ages
Quickfire TV Quiz: Challenge a friend, or a computer
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 Strategies
The questions surrounding AMÁLIA reflect broader issues faced by European nations in developing sovereign AI models. Openness determines transparency and collaboration potential; native-language data volume influences model performance and cultural relevance; and clear optimization goals shape strategic outcomes. Addressing these questions is crucial for Europe’s leadership in AI and for ensuring that investments translate into effective, autonomous AI capabilities that serve national interests.
European Sovereign-Language Model Initiatives and Challenges
Across Europe, countries like Italy, Germany, France, and Norway are investing in their own large language models, often with similar structural questions about openness, native data, and goals. The European Union’s open consortium, OpenEuroLLM, exemplifies collective efforts to build sovereign models. These initiatives are driven by strategic concerns over dependence on foreign AI giants and the desire for culturally and linguistically tailored AI tools. However, many projects, including AMÁLIA, are still navigating fundamental questions about how open they truly are, how much native data is enough, and what their primary objectives should be—issues that influence the effectiveness and strategic value of these models.
“The core questions about openness, native-language data, and goals are not just technical; they are strategic, shaping the future of European AI sovereignty.”
— Duarte O. Carmo
Unresolved Questions About AMÁLIA’s Strategic and Technical Limits
It remains unclear how open AMÁLIA truly is in practice, given the limited native Portuguese data and the model’s current capabilities. The final version scheduled for June 2026 may address some gaps, but the extent to which these issues will be resolved is still uncertain. Additionally, the primary strategic goals of the project—whether to prioritize openness, native-language performance, or broader AI autonomy—are not yet explicitly defined or publicly clarified.
Upcoming Milestones and Further Evaluations of AMÁLIA
The next 12-24 months will be critical for evaluating AMÁLIA’s development. The final version due in June 2026 is expected to clarify some of the current uncertainties, potentially expanding native-language capabilities and increasing transparency about openness. Meanwhile, ongoing critical analysis and comparisons with other European models will continue to shape the discourse around the project’s strategic direction and effectiveness.
Key Questions
What are the main concerns about AMÁLIA’s openness?
Experts question whether AMÁLIA’s current architecture and training data truly reflect an open model, especially given the limited native Portuguese data and the continuation of a pre-existing multilingual foundation rather than a from-scratch approach.
How much native Portuguese data does AMÁLIA use?
Approximately 5.8 billion tokens from Portuguese web archives, representing about 5.5% of the total training data, which may be insufficient for some performance benchmarks.
What are the strategic goals of AMÁLIA?
It is not yet clear whether the project aims primarily for native-language excellence, openness, or broader AI autonomy, as these priorities have not been explicitly communicated by the developers or the government.
Will the final version address current limitations?
It is expected that the June 2026 release will improve native Portuguese capabilities and possibly clarify the project’s openness and strategic aims, but this remains to be seen.
Why do these questions matter for European AI development?
Addressing these questions is vital for Europe’s strategic independence in AI, ensuring that investments translate into effective, culturally relevant, and transparent models that can compete globally.
Source: ThorstenMeyerAI.com
