Kyutai is setting out to make open research central to its AI work. The Paris-based lab has a budget of nearly €300 million, or $330 million, and says it intends to publish not only its models but also the code and documentation behind their training.
A large budget and a research team
Xavier Niel, CEO of Iliad, gave an update on Kyutai’s funding at the ai-PULSE conference. Contributions have come from Rodolphe Saadé, CEO of CMA CGM, the Eric Schmidt Foundation, and others. The lab is privately funded.
The conference was organized by Scaleway, Iliad’s cloud division. Niel said Scaleway will provide Nvidia H100 GPUs for Kyutai’s research projects at cost. That support connects the lab’s funding to computing resources it can use for its research.
Kyutai has started hiring its core scientific team. The group includes former researchers from Meta’s FAIR and Google’s DeepMind. Patrick Perez, formerly of Valeo, will lead the lab. Its scientific advisors include Meta’s Head of AI Yann LeCun and German Leibniz Prize winner and AI researcher Bernhard Schölkopf.
Open publication is part of the plan
Kyutai says all its models will be open source. The lab also plans to share training source code and documentation describing how models were trained. That commitment could let other researchers examine how the work was done, alongside the models themselves.
Alexandre Defossez, a member of the core team, said the lab is not necessarily aiming to move as quickly as Mistral. He described its ambition as providing scientific purpose, understanding, and a code base to explain its results. The emphasis, then, is on making the research behind a release available as well as the release itself.
Kyutai hopes to publish its first model within a year. The source describes this as a target, while making clear that the lab’s approach gives explanation and documentation a central place in its work.
France’s argument about AI rules
French President Emmanuel Macron voiced support for open source at the conference and said it should become a strength for France. He also defended France’s position on the European AI Act, arguing that regulation should focus on use cases rather than model makers.
France has been lobbying to weaken the law during trilogue negotiations among the Parliament, the Commission, and the Council. Niel backed France’s position, saying Europe is lagging in AI innovation and that regulation could slow new European entrants and reduce their chances of catching up.
The conference therefore placed Kyutai’s research plans alongside a broader debate about how Europe can support AI development. The lab’s stated approach is to make its research open and explainable; the policy argument presented by Macron and Niel is that regulation should not impede innovation by European newcomers.
What the lab’s approach signals
Kyutai’s combination of private funding, research hires, computing support, and a commitment to publication gives the lab a defined starting point. Its first model will offer a concrete opportunity to see how fully it carries out its plan to share both models and the materials behind their training.
For now, the central promise is transparency around research. Kyutai’s timetable and the scientific resources described at the conference indicate its intended direction, while the planned publications will show what that looks like in practice.