A single social-media post claims OpenAI released 722 mathematical manuscripts from an unreleased model, but no OpenAI statement in the provided material confirms the release, manuscript count, or model status.
The claim originates from the X account @RussellQuantum, which posted on 10 October that "OpenAI released 722 manuscripts this week from an unreleased model - algebra, number theory, theoretical computer science, logic, topology" Signals — launches and leaks. The post also said the results covered "372 open math problems" with "an average of three hours of compute each."
Automatica reported earlier on 10 October that OpenAI had published "719 mathematical results generated by internal AI model," citing mathematicians who told The Verge the field would need years to digest the work Automatica Press. That reporting describes a related OpenAI math release but does not independently corroborate the newer claim of 722 manuscripts from an unreleased model.
The number mismatch is one reason for caution: the earlier coverage references 719 mathematical results, while the social post claims 722 manuscripts Automatica Press Signals — launches and leaks. The material provided does not clarify whether the difference reflects revised counting, different document types, or a mistaken claim.
The same post makes further assertions that Automatica has not independently verified. It states that "OpenAI's own advisory group, formed after the scrutiny over its Navier–Stokes claim, called the release "the beginning, not the completion" of understanding" and that "only a fraction of the results are formalized in Lean" Signals — launches and leaks. The post also references reactions from Ethereum Foundation researcher Justin Drake and others, but no source beyond the social post supports those statements in the provided material.
A public OpenAI release, repository, paper set, or statement matching the manuscript count, scope, and model status would confirm the basics; an OpenAI denial or materially different official record would undermine it. No such OpenAI statement appears in the material provided.