INRIA

Inria (France's National Institute for Research in Digital Science and Technology) is the anchor of French academic AI. It runs its research as joint project-teams embedded inside the universities, so a single Inria banner captures most of the cluster: SIERRA (with ENS Paris/PSL — Francis Bach, optimization/ML theory), ALMAnaCH (Benoît Sagot — the CamemBERT/Gaperon LLM line), Soda (Gaël Varoquaux — scikit-learn stewardship + tabular FMs), WILLOW (vision/robotics with DI-ENS), and ARAMIS (neuroimaging, at the Paris Brain Institute). The pure ENS/PSL crown jewels — the LPENS physics-of-learning group and Gabriel Peyré's optimal-transport line — attach as the intertwined academic partner, formalized by the €75M PR[AI]RIE-PSAI institute (CNRS + Inria + Pasteur + PSL + Paris Cité, launched June 2025).

Tracker-relevant output spans four axes at once: a top-tier theory result (the NeurIPS 2025 Best Paper on why diffusion models don't memorize, from ENS-PSL's LPENS), a genuinely shipped open-model line (CamemBERT → Gaperon), scientific tabular foundation models (TabICL, CARTE — the European counterweight to TabPFN), and the world's most-adopted ML library, scikit-learn (66K+★, now stewarded via the :probabl. spinout).

The cluster is a heavy talent-donor — Meta FAIR Paris, Kyutai, Mistral, Hugging Face, Owkin, and Apple's Paris ML lab all draw from it, and Cordelia Schmid's headline vision work is Google-branded (excluded here). Lucie/Luciole (LINAGORA/OpenLLM-France) is a separate government-backed consortium, not Inria.