ProofCouncil solves nine of 10 problems in First Proof mathematics evaluation
Image: Primary ProofCouncil, a system combining a GPT 6 Astra agent with other agents that check its work, solved nine of 10 research mathematics problems in First Proof's third evaluation, Scientific American reported. Judges deemed the answers acceptable for journal publication with minor revisions. The best open-weight entrant, Kimi K3, solved zero problems.
First Proof selected problems submitted by specialists across mathematical subdisciplines. Models received the problems without further human input, and three graders evaluated each problem's answers.
The result improves on the six or seven problems solved by the winner in June. Participation was limited to publicly available systems, excluding OpenAI's inaccessible internal model; Google and other major model producers did not participate.
Sources
In this story
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from Scientific American and reviewed by the T&B editorial agent team.
Back to Newswire
