Google DeepMind's AI solves dozens of open math problems, and a top journal published it

Google DeepMind says its AI solved nine long-open math problems and proved dozens of other unproven claims. Outside experts reviewed the work before it was published.

The Decoder reports that AlphaProof Nexus solved nine of 353 open problems on a well-known list called the Erdős problems. Two of those had gone unsolved for 56 years.

The system also proved 44 unproven claims from a large database of number patterns. It solved an open question in algebraic geometry too, and it improved a known limit in a math technique used to train some AI models.

Here is how it works. Google's Gemini 3.1 Pro model writes and rewrites a proof. A checking program then tests every draft, and its error messages guide the next try. For harder pieces, the system can call on a separate proof tool called AlphaProof. Google DeepMind says it solved the nine Erdős problems even without that tool.

The work comes during a heated debate. OpenAI recently posted hundreds of AI-written proofs online instead of sending them to journals. DeepMind chose peer review.

The facts

• AlphaProof Nexus solved nine of 353 open Erdős problems, two of which had been unsolved for 56 years.
• It proved 44 conjectures from the OEIS integer sequence database.
• It also solved an open question in algebraic geometry and improved a known bound in min-max optimization.
• The system is built around Gemini 3.1 Pro, which drafts and revises proofs in the formal language Lean, and Lean checks each draft after every edit.
• Google DeepMind reports it solved the nine Erdős problems even without the specialized AlphaProof tool.
• The results were published in a study in Science, and the team shared the solutions online ahead of the paper.
• OpenAI recently put hundreds of AI proofs on GitHub instead of submitting them to journals.

Sources

• The Decoder: Google Deepmind's AlphaProof Nexus solved dozens of open math problems and actually got peer reviewed