DeepMind's multi-agent systems, backed by strict proof checkers, are chipping away at problems that have resisted human mathematicians for decades
Google’s Gemini-powered math agents have found proofs for five math problems, using a team of AI agents whose work gets graded by unforgiving automated checkers.
The best-documented version of this approach is a framework DeepMind calls AlphaProof Nexus. It runs multiple independent prover subagents in multi-turn reasoning loops, powered by Gemini 3.1 Pro.
The agents don’t fire off one answer and walk away. They reason, revise, and try again over several rounds until a proof either checks out or doesn’t.
The numbers behind the headline #
AlphaProof Nexus resolved 9 out of 353 previously unsolved Erdős problems. Two of those nine had stumped mathematicians for 56 years.
The same framework proved 44 out of 492 conjectures listed in the Online Encyclopedia of Integer Sequences, known as OEIS.
DeepMind’s results came at a computational cost of only a few hundred dollars per problem.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
A detailed arXiv paper describing the work was posted around mid-2026. Domain experts reviewed the results and confirmed the proofs, along with the faithful formalization of the original conjectures.
Formalization means translating a human-written math problem into precise machine-checkable language, and a sloppy translation can mean proving the wrong thing entirely.
Aletheia and the Olympiad run #
A separate system called Aletheia, built on Gemini Deep Think, operates as a semi-autonomous research assistant. Since its deployment in December 2025, Aletheia has evaluated approximately 700 open Erdős problems. It identified or created solutions for 13 of them, earning praise from expert reviewers. Four of those 13 were novel autonomous resolutions.
Gemini Deep Think also has a competition track record. At the 2025 International Mathematical Olympiad, it scored 35 out of 42 points, reaching gold-medal standard. It solved 5 out of 6 problems within the standard time limits that human contestants face.
Why the agentic approach is getting attention #
Pushmeet Kohli and other DeepMind researchers have emphasized that this agentic approach could extend beyond pure math. They pointed to fields such as combinatorics and quantum optics as potential targets. The research also highlights applications in algebraic geometry.
What this means for mathematics and AI #
The research notes that current AI systems still require human oversight for novelty assessment, meaning people must confirm whether a result is genuinely new.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our