OpenAI has published the full output of an unreleased internal model turned loose on open mathematical problems. Title: openai/math URL: ❓ What exactly was released? 722 manuscripts organized into 372 result families. Roughly 4,000 problems were posed to the model across the evaluation period, and the vast majority of results came out of the same procedure. It's Apache-2.0, with PDFs, source files, and citation info for each manuscript. ❓ Why release it? Because performance on existing benchmarks had plateaued. Once problems with known answers stopped being informative, they extended evaluation to open research questions. Each result consumed about three hours of thinking compute on average. ❓ Is the math guaranteed correct? No, and this is the part that matters most. The README states plainly that the collection holds results at different stages of verification, that not all of them have accompanying Lean formalizations, and that some unformalized results could have issues. The index marks formalized results with a Lean link so you can tell which ones are machine-checked. ❓ So the Riemann Hypothesis is solved? No. The two results singled out as exceptions are a zero-free region for the zeta function, which is a partial result, and a proof of the Hodge Conjecture restricted to CM abelian varieties, a specific class. Neither settles the conjecture itself, and the zeta writeup was human-edited for readability. #Mathematics# #OpenAI#