First Proof: Third Batch Benchmark Report (1stproof.org)

The Third Batch Benchmark has been announced to evaluate AI systems' performance on ten previously unsolved mathematical problems, which human mathematicians have tackled but not publicly shared. This initiative invites mathematicians to submit unpublished problems with concise proofs, undergoing a rigorous selection process before being tested. The benchmark aims to push the boundaries of AI in handling complex mathematical inquiries by limiting interaction; each AI system gets only one attempt to answer the problems presented. This project is significant for the AI/ML community as it not only challenges current AI capabilities but also fosters collaboration between mathematicians and AI researchers. The results will be graded by approximately 30 expert referees who will assess the AI-generated solutions for correctness, clarity, and proper attribution. The publication of findings, scheduled for October 14, 2026, promises to provide invaluable insights into how AI systems can generate mathematical proofs, potentially paving the way for advancements in automated reasoning and theorem proving.