AI Co-Mathematician

"AI Co-Mathematician: Accelerating Mathematicians with Agentic AI." Agentic-AI workbench from Google DeepMind that lets working mathematicians interleave AI agents into ideation, literature search, computational exploration, theorem proving, and theory building — backed by an asynchronous, stateful workspace coordinated by a top-level project coordinator across parallel research workstreams.

Sets a new SOTA of 48% on FrontierMath Tier 4, the Epoch AI benchmark of 50 research-level problems designed to surpass Tier 3 in difficulty. In a documented use case, Oxford's Marc Lackenby used the system to resolve Problem 21.10 from the Kourovka Notebook in group theory after a reviewer agent flagged a flaw in the first proof attempt.

By Daniel Zheng, Ingrid von Glehn, Yori Zwols, Lars Buesing, Martin Wattenberg, Alex Davies, Pushmeet Kohli, and collaborators (Google DeepMind).