Fabian Gloeckle

Fabian
Gloeckle

Fabian Gloeckle Portrait

I am a machine learning researcher and mathematician working on language models for code and math. I am completing my PhD at Meta FAIR Paris and Institut Polytechnique de Paris, advised by Amaury Hayat and Gabriel Synnaeve.

In my PhD, I developed multi-token prediction (MTP), a pretraining objective now used in leading models such as DeepSeek-V3, DeepSeek-V4, and NVIDIA Nemotron Ultra, jointly with Badr Youbi Idrissi. I also pursued explorative research into reinforcement learning for multi-agent and hierarchical inference. On formal mathematics, I was the first to fully automatically formalize an entire textbook in Lean, and I built an AlphaProof-style MCTS prover for Lean from scratch.

To advance automated scientific research, we must build highly parallel AI systems that interact with verifiers, self-verify, and conduct open-ended exploration.

Email: firstname dot lastname at enpc dot fr

Fabian Gloeckle Portrait

Updates

  • Jun 2026 Joining a stealth startup.
  • Jun 2026 Gave a talk at the 5th workshop on Mathematics and AI (AI & Mathematics in the Netherlands), University of Groningen, Netherlands.
  • May 2026 Advancing autoformalization with Formal Frontier and the Mathlib Initiative.
  • Mar 2026 Finished my PhD contract at FAIR; now on the industry research job market.
  • Mar 2026 Released ABEL open-source.
  • Dec 2025 Gave a talk at the AI Morning for Mathematics, Laboratoire Jacques-Louis Lions (LJLL), Sorbonne Université, Paris.
  • Nov 2025 Gave a talk at Timothy Gowers' seminar, La philosophie de la pratique des mathématiques, Collège de France, Paris.
  • Oct 2025 Recognized as a Top Reviewer for NeurIPS 2025.
  • Sep 2025 Gave a talk at the Machine Learning and AI for Mathematics workshop, Mathematisches Forschungsinstitut Oberwolfach (MFO), Germany.
  • Sep 2025 Gave a talk at the Mechanization and Mathematical Research workshop, Lorentz Center, Leiden, Netherlands.
  • Jul 2025 Gave a talk at the Machine Learning and Mathematics workshop, Korea Institute for Advanced Study (KIAS), Seoul.
  • May 2025 Gave a talk at the Mathematics for and by Large Language Models workshop, Institut des Hautes Études Scientifiques (IHES), Bures-sur-Yvette.
  • Apr 2025 Gave a talk at the Deep Learning Models for Mathematics and Type Theory workshop, Chalmers AI Research Centre (CHAIR), Gothenburg, Sweden.
  • Jul 2024 Gave a talk at the Sacl-AI 4 Science Workshop, Palaiseau.
  • Sep 2023 Gave a talk at the Congrès des Jeunes Chercheurs en Mathématiques et Applications (CJC-MA), CentraleSupélec, Gif-sur-Yvette.

Selected Publications

Loading publications...

See all publications on Google Scholar →