Prateek Tanmay
3 min readNew DelhiUpdated: May 26, 2026 04:55 PM IST
Google DeepMind has claimed that its new AI agent, AlphaProof Nexus, autonomously solved nine out of 353 open Erdos problems in a purported milestone related to AI and mathematics.
First posed by legendary mathematician Paul Erdos in 1946, some of the problems had remained unsolved for the past 56 years. However, DeepMind researchers said that AlphaProof Nexus solved nine of these problems while relying on only a few hundred dollars-worth of computing resources.
AlphaProof Nexus also proved 44 open OEIS conjectures, resolved a 15-year-old question in algebraic geometry, and even discovered a new algorithmic parameter in optimisation theory that humans had not previously found, the researchers further claimed in a study titled ‘Advancing Mathematics Research with AI-Driven Formal Proof Search’ published on May 21.
Google’s latest mathematical achievement comes just days after OpenAI said one of its AI models had cracked the famous “planar unit distance problem”, which was unsolved for the last 80 years.
AI models have increasingly shown progress in mathematical capabilities over the past few years. In 2024, researchers from Google DeepMind introduced Alpha Geometry, an AI model capable of solving complex geometry problems at a level comparable to International Mathematical Olympiad contestants.
OpenAI and Google DeepMind’s unreleased models posted similar results at the International Math Olympiad in 2025. However, OpenAI has also sparked controversy surrounding its mathematical claims in the past.
Meanwhile, Google’s latest work with AlphaProof Nexus potentially stands out because the AI agent autonomously generated the proofs, which were reportedly verified with AI tools instead of relying entirely on human mathematicians.
Story continues below this ad
The researchers said that a combination of AI-generated proofs and formal verification could change how mathematicians work in the future. “Our results support this vision. Our mathematician collaborators found that proof attempts by our agents enhanced their understanding of a problem, even when an agent could not prove the claim at hand,” they wrote.
With regards to OpenAI’s earlier claims, the ChatGPT-maker said it relied on mathematicians Noga Alon, Thomas Bloom, and Tim Gowers, to review and comment on the AI-generated findings.
However, Google researchers took a different approach. They combined an AI reasoning model, likely Gemini 3.1 Pro, with a formal verification system called Lean. While the model generated proofs, the Lean system automatically checked each logical step based on strict mathematical rules.
© IE Online Media Services Pvt Ltd
