On Sunday, September 21, 2026, OpenAI made one of the boldest research claims of the AI era: an internal model has resolved more than 100 previously open mathematics problems , on top of the Navier-Stokes Millennium Prize proof it published earlier this month.
Also Read: IBM and NASA Open-Source an AI Model of the Moon - News
At the same time, the company announced an independent Advisory Group on Mathematics and Artificial Intelligence, hosted at Princeton's Institute for Advanced Study (IAS), to help review and release the results.
TL;DR
An unreleased OpenAI model, which started training on August 28, is behind the results
OpenAI says it has resolved 100+ open problems across most areas of mathematics
The same model produced OpenAI's claimed Navier-Stokes solution, published with a Lean formalisation
Most of the 100+ results haven't been released yet, so outsiders can't verify them
A nine-member advisory group, including Timothy Gowers, Edward Witten and Martin Hairer, will advise on review and publication, but not on how fast OpenAI moves
Which areas of maths?
According to OpenAI's announcement as reported by Startup Fortune, the problems span high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, operator algebras, quantum complexity, lattice cryptography and extremal combinatorics.
Also Read: Meta Launches Muse, an AI Agent That Acts for You
OpenAI said the pace of progress surprised its own in-house mathematicians.
Who's on the advisory group?
The nine founding members are:
François Charles
Camillo De Lellis
Timothy Gowers
Martin Hairer
Nikhil Srivastava
Ulrike Tillmann
Ravi Vakil
Edward Witten
Melanie Matchett Wood
Four of them hold Fields Medals. Members are unpaid by OpenAI, can change the group's membership themselves, may give advice OpenAI hasn't asked for, and are free to publish their own conclusions. Their first job is advising on how to coordinate the release of a large backlog of results.
Also Read: Programming: Shopify Ditches React
The catch: advice, not brakes
OpenAI was explicit that the group won't advise on the pace of its internal mathematical research. IAS echoed this in its own statement, stressing it has no decision-making power at any AI company.
That matters because the announcement follows real friction. Earlier this month, 25 Fields Medal winners signed an open letter — titled "A Severe Misalignment of AI in Mathematics" — arguing that AI labs racing to claim famous problems as benchmarks threatens mathematicians' work. TechCrunch noted that only one advisory group member, Camillo De Lellis, also signed that letter.
Also Read: Laravel: Carbon Immutable vs
Why this matters beyond maths
The Neuron framed the real problem neatly: if AI can produce research faster than humans can check it, the bottleneck shifts from discovery to verification . A nine-person volunteer panel reviewing 100+ proofs across many fields is a tall order on any human timescale.
For developers and tech watchers, a few takeaways:
Formal verification is becoming central. Publishing proofs in Lean, a proof assistant, lets machines check machine-generated work.
Capability jumps are fast. The model began training less than a month before the announcement.
Credit and attribution are unresolved. Who gets credit for an AI-generated proof is still an open question.
What's next
Expect results to be released in batches, coordinated with the advisory group. Watch for independent verification — until mathematicians outside OpenAI confirm the proofs, treat the "100+" figure as a claim, not a fact.
Also Read: News: Meta Muse Zero-Day:
FAQ
Did OpenAI really solve Navier-Stokes? OpenAI published a write-up and Lean-formalised proof earlier in September 2026. The claim has drawn scrutiny, and independent evaluation is ongoing.
Which model solved the problems? An unnamed internal OpenAI model that began training on August 28, 2026. It isn't publicly available.
Also Read: Grok 4.7 Released: xAI's New Model Keeps 2/6 Pricing | Benchmarks & How It Compares
Where is the advisory group based? The Institute for Advanced Study in Princeton, New Jersey.
Sources: TechCrunch, The Daily Star, The Neuron, Business Standard, Startup Fortune, AI Weekly.
