OpenAI publishes 722 math manuscripts from an unreleased model — Lean-checked proofs, ~3 hours of compute each
OpenAI released 722 mathematical manuscripts in 372 families from an internal AI model that has not been released, most with machine-checkable Lean proofs.
What happened
On October 6, 2026, OpenAI published 722 mathematical manuscripts — grouped into 372 families of related results — in a public GitHub repository. The company says the results were produced by an internal AI model that has not been released.
What's in the release
Many of the proofs ship with versions written in Lean, a language that lets a computer check a proof, though not all of them do yet. The release also includes reasoning summaries for 10 of the results, compute estimates, and attempted-problem statistics: the model was given about 4,000 problems, and each result used the equivalent of roughly three hours of ChatGPT Pro thinking on average.
Why it matters
The release landed days after the Advisory Group on Mathematics and Artificial Intelligence published recommendations on September 29, 2026 asking AI labs to stop testing advanced math problems on proprietary models, and to publish prompts, reasoning summaries, and compute costs for results the prompters don't yet fully understand. OpenAI released reasoning summaries and compute statistics but did not publish the prompts behind the results.
Sources
- https://aistockwire.com/blog/openai-722-math-manuscripts-unreleased-model-advisory-group-october-2026
- https://www.scriptbyai.com/timeline-of-chatgpt/
- https://willreset.com/tibo
Verified October 6, 2026.
Sources
Get updates like this every morning
- ② Card on Stripe
- ③ 7 days free
Then $2/month · cancel anytime in one click