OpenAI Model Solves Thousands of Math Problems

OpenAI's internal model demonstrated outstanding results in the field of mathematics, solving thousands of complex tasks within a limited time.

Model Results

The company published the testing results of its unnamed internal neural network. The system gained access to a database of approximately 4000 mathematical problems of varying complexity.

The outcome of the computations was the generation of 722 manuscripts containing proofs or solutions. These materials cover 372 different scientific directions and topics.

The average time spent to obtain one result was about three hours. Computations were performed at the performance level of ChatGPT Pro.

Community Verification and Reaction

The published results are currently being verified by leading specialists worldwide. Mathematicians express skepticism, noting that full-fledged scientific publications require detailed articles rather than brief reports.

Access to Source Data

All resulting manuscripts and formalizations of proofs are available on GitHub. Lean formalizations of some presented solutions for independent verification are also posted there.