The European Commission has chosen a consortium led by the Milan-based artificial intelligence startup Domyn to build a large open-source frontier AI model for Europe, a move intended to give the bloc a homegrown alternative to systems from OpenAI, Anthropic, and Chinese developers. The project, carried by a grouping named the EUROPA consortium, calls for a model of roughly 400 billion parameters whose open weights would be made available to European researchers and businesses, covering all 24 official languages of the European Union.
Reuters reported that Domyn's chief executive expects the model to be released within a year, framing it as the centerpiece of Europe's push for "sovereign" AI. The decision, confirmed in reporting across outlets including Il Sole 24 Ore, the Brussels Times, and Anadolu Ajansi, marks the first time the EU has put a single consortium in charge of building a flagship open model under its broader AI competitiveness strategy. For more on the continent's ambitions, follow our latest AI developments.
Why Europe Wants Its Own Frontier Model
European policymakers have grown increasingly uneasy about the bloc's dependence on a handful of US companies for the most capable AI systems, and on Chinese providers that have rapidly gained global usage. The concern is not only commercial. EU officials have repeatedly argued that models trained and controlled outside Europe may not reflect European values, privacy standards, or regulatory expectations, and that strategic sectors from public administration to healthcare should not sit on infrastructure owned by foreign vendors.
The Domyn-led effort is the most concrete answer yet to that anxiety. Rather than waiting for European labs to close the gap with US frontier models on their own, the Commission is effectively commissioning one, using public procurement and a competitive challenge to assemble the talent and compute needed to train a model that European governments can trust.
A 400-Billion-Parameter, Multilingual Model
The specification is ambitious by any standard. A 400-billion-parameter model would place the EUROPA system among the largest open-weight models in the world, in the same weight class as the biggest openly released systems to date. Equally important is the language requirement: the consortium is tasked with delivering strong performance across all 24 official EU languages, from Spanish and French to Finnish, Irish, and Maltese.
Multilingual coverage has long been a weak spot for models trained predominantly on English data. By making 24-language quality a core deliverable rather than an afterthought, the project aims to produce a model that is genuinely useful for European public services and businesses that operate in languages US-centric tools handle poorly.
The Frontier AI Grand Challenge
The selection flowed from the EU's Frontier AI Grand Challenge, a competitive process designed to identify a European team capable of training a frontier-class model. Il Sole 24 Ore described the result as Domyn taking the lead on what officials have cast as a sovereign AI project. The challenge sits alongside the bloc's wider investments in AI factories and compute infrastructure, which together are meant to give European developers the raw power needed to compete at the frontier without renting it from American cloud giants.
Details on the full roster of consortium partners and the total budget envelope have trickled out gradually, but the thrust is clear: the EU wants a coordinated, publicly backed build rather than a fragmented landscape of smaller national efforts.
The Milan Startup at the Center
Domyn, an Italian AI company based in Milan, has until now been little known outside specialist circles. Startup Fortune noted that Europe is effectively "betting its AI sovereignty on a Milan startup most people have never heard of," reflecting both the ambition of the bet and the relative obscurity of the company entrusted with it. The selection elevates Domyn into the front rank of European AI labs overnight, with all the scrutiny that entails.
For Domyn, the assignment is a defining test: delivering a frontier-grade, multilingual, open model within roughly a year would validate the Commission's confidence and establish Italy as a serious AI hub. Missing that target would hand critics evidence that Europe cannot match the pace of US and Chinese labs even with direct state backing.
The Open-Weights Bet
The decision to release the model as open weights is itself a strategic choice. Open weights allow independent auditing, academic research, and downstream fine-tuning by European companies, addressing calls for transparency that closed US systems do not satisfy. It also aligns with the EU's regulatory philosophy under the AI Act, which favors scrutinizable systems.
The trade-off is competitive exposure: publishing weights lets rivals study and build on European work. EU planners have accepted that risk, betting that a thriving open European ecosystem will ultimately matter more than keeping a single model proprietary. Whether a 400-billion-parameter, 24-language model delivered within a year is enough to close the gap with the front of the pack will be the central question for European AI in 2026.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
