According to Decrypt, OpenAI posted 722 manuscripts written by an unreleased AI model, stating that most were generated from a single prompt. The company described the output as comprising hundreds of open math problems solved through that prompt.
Mathematicians referenced in the report expressed uncertainty about the claims, describing them as unverified. They requested official documentation—such as the exact prompt used, model version details, and verification steps—to assess the results independently.
No additional information about timing, testing methods, or internal validation processes was provided by OpenAI in the publicly shared material. The manuscripts themselves were made available without annotation regarding provenance or verification status.
