On Oct 6 (US Eastern time), OpenAI published 722 math research manuscripts on GitHub written by an unreleased internal model, grouped into 372 result categories across 17 fields, with 162 carrying Lean formalizations of their main results; the company said each successful result cost roughly the equivalent of ChatGPT Pro thinking for three hours, and listed targets such as the four-dimensional Kakeya conjecture, the Unique Games conjecture and Mahler's conjecture. The release drew heavy skepticism, coming after OpenAI's Sept 8 claim to have solved the Navier–Stokes Millennium Prize problem; a Milan-Bicocca researcher says a recent paper finds at least two inconsistencies between the paper and its Lean code, making the formalized result weaker. OpenAI has released only average compute times, not the model, prompts or full computation traces, and has withdrawn 3 manuscripts and revised 14 after finding a symbol error, leaving 719 listed.