A competition, not yet a model
On 19 June 2026, the European Commission did not "launch" an artificial intelligence model: it designated the winner of a competition tasked with building one. The EUROPA consortium, led by the Italian company Domyn with the German institutes Fraunhofer IAIS and Fraunhofer IIS, won the Frontier AI Grand Challenge. Its mission: to develop an open source model described as "frontier," meaning at the level of the most advanced systems, covering the 24 official languages of the Union.
The competition had been launched on 13 February 2026 by the Commission, in connection with the EuroHPC Joint Undertaking and within the framework of the AI-BOOST project. It required a model of at least 400 billion parameters, encouraging resource-efficient "mixture of experts" architectures. Only one project was to be selected. The prize is not a traditional subsidy: the winner receives up to 2,5 % of EuroHPC's total computing capacity for one year, on European supercomputers optimized for AI.
The Executive Vice-President for Technological Sovereignty, Henna Virkkunen, presented this choice as a political gesture:
"Europe can be at the forefront of advanced AI on its own terms. EUROPA will build a European frontier AI model in the 24 EU languages, showing that we can match the best while remaining true to our values." (Henna Virkkunen, Commission press release, translated from English)
What EUROPA promises
According to AI-BOOST, which coordinates the competition, the consortium aims for a model of 500 billion parameters, trained exclusively on open data and published under the Apache 2.0 license, so that any European institution can audit, reproduce, and extend it. The project also promises to better integrate Europe's own legal, administrative, and cultural knowledge, often poorly covered by models designed elsewhere.
The bet is twofold. First, it involves demonstrating that a very large-scale model can be trained on European public infrastructure, which Domyn's CEO, Uljan Sharka, presents as proof that "Europe can develop, control, and continuously advance its own frontier AI capabilities." Second, it involves offering total transparency, from data to license, which would distinguish EUROPA from many "open-weight" models whose training data remain secret.
Domyn, a winner with international ties
The choice of the winner deserves attention. Domyn, formerly iGenius, is a Milan-based company specializing in AI for regulated sectors: finance, administration, industry. It is building in Italy the supercomputer Colosseum, presented as one of the largest in Europe, in partnership with Nvidia and the Emirati group G42. In May 2025, G42 announced, via its subsidiary Core42, a strategic partnership with iGenius to deploy this computing cluster and ensure its operation on a European scale.
This does not call into question the European character of the consortium or the promised openness of the model. But it reminds us that "sovereignty" in AI is rarely pure: capital, chips, and industrial partners cross borders. The future governance of the model, its updates, and its training infrastructure will therefore be a point to watch.
A galaxy of European initiatives
EUROPA is not starting from scratch. On 16 July 2026, the European Commission's Directorate-General for Translation published the EU Institutional LLM, a model built from an existing European open model and enriched with multilingual data from the institutions. It is downloadable by any legal entity established in the Union. On the private side, Mistral AI unveiled in October 2026 Mistral Large 4, an open-weight model with 1 000 billion parameters, trained on more than 160 languages, including all those of the Union.
EUROPA's specificity therefore lies less in multilingualism than in its status: a frontier model funded by public authorities, entirely open and reproducible. The risk is that of dispersion. Between research projects, institutional models and private champions, Europe is multiplying initiatives without always clarifying how they fit together or the uses that should result from them.
The questions that remain open
- The timeline. The model must be developed within one year of computing access. Yet the technological frontier moves quickly: a model with 500 billion parameters might no longer be "frontier" by the time it is released.
- The resources. A share of EuroHPC's capacity for one year remains modest compared to the spending of American and Chinese private labs, even if the goal is not to match them but to prove capability.
- The usage. A public model only makes sense if it is adopted. Will public procurement, administrations and European companies use it, or will they continue to turn to already integrated foreign solutions?
- Compliance. As a general-purpose AI model, EUROPA will be subject to the obligations of the AI Act, which came into force on 1 August 2024 and whose application was partially revised by the omnibus adopted in June 2026.
A test for European strategy
With EUROPA, Brussels is changing its stance: after primarily regulating AI, it now seeks to produce it. The initiative has the merit of being consistent with the discourse on strategic autonomy, and its choice of full openness could make it a common good useful to researchers, administrations and SMEs. But its success will be measured neither by the number of parameters nor by announcements: it will depend on the actual delivery of the model, its real quality and its adoption. See you in a year.
Sources: European Commission, "Commission selects EUROPA consortium as the winner of the Frontier AI Grand Challenge", 19 June 2026 ; EuroHPC JU, launch of the Frontier AI Grand Challenge, 16 February 2026 ; AI-BOOST, "Frontier AI Grand Challenge: The Winner is Out" ; Domyn, press release of 19 June 2026 ; G42, partnership with iGenius (now Domyn), 20 May 2025 ; DG Translation, "EU Institutional LLM released as an open model", 16 July 2026 ; Mistral AI, "Introducing Mistral Large 4", 6 October 2026 ; Council of the EU, adoption of the AI omnibus, 29 June 2026.
