Google Deepmind released AlphaProof Nexus on October 12, 2026, saying it solved dozens of open math problems on its own. It is the company's first major research breakthrough since its earlier work on AI proof systems.
The team reported solving 44 conjectures from the OEIS integer sequence database and nine of 353 open Erdős problems, two of which remained unsolved for 56 years. That compares with previous efforts that often failed to achieve peer review.
AlphaProof Nexus is built on Gemini 3.1 Pro, which drafts and revises proofs in the formal language Lean. It targets complex mathematical problem-solving and is available to researchers through online platforms.
"Lean checks each draft after every edit, and error messages guide the next attempt," said Manuel Uth, the study's lead author. The system can also call on AlphaProof, a specialized proof tool, for harder sub-problems.
The announcement follows a heated debate over AI math, with OpenAI recently dumping hundreds of AI proofs on GitHub instead of submitting them to journals. Deepmind put its work through peer review, marking a significant shift in the field.
Google Deepmind did not say how it plans to expand the system's capabilities and raised the open question of whether AI can replace human mathematicians. The study highlights the potential for AI to contribute to mathematical research.
Source: thedecoder
