OpenAI stated that an internal model solved more than 100 open mathematics problems after approximately 24 days of training. The announcement was released on the company's blog on September 21, alongside the creation of an independent advisory group at the Institute for Advanced Study in Princeton.
What Happened?
Training for this internal model began on August 28. According to OpenAI, it covered most areas of mathematics and includes the previously disclosed result on the Navier-Stokes Millennium problem. The company has not published the list of over 100 open math problems or their corresponding proofs.
Regarding Navier-Stokes, announced in early September, OpenAI described an unreleased model, thousands of AI agents, and dozens of compute hours to argue that fluid equations can "blow up" in finite time. This proof remains under debate.
What Does This Mean?
"Solved" in this announcement is a company claim—not a stamp of approval from the Clay Mathematics Institute. Without a public index, it is impossible to know how many of the open math problems are complete proofs, how many are partial advances, and how many still require formal human checking, for instance in Lean.
This is why the Advisory Group on Mathematics and Artificial Intelligence was created. The initial group includes Timothy Gowers, Martin Hairer, Edward Witten, Melanie Matchett Wood, François Charles, Camillo De Lellis, Nikhil Srivastava, Ulrike Tillmann, and Ravi Vakil. They can publish recommendations and speak publicly, but they do not veto OpenAI's internal schedule.
Why Is This Important?
If part of these results are confirmed, mathematical research faces a new bottleneck: AI produces proof candidates faster than academia can read, teach, and formalize them. Outlets like TechCrunch and 36Kr treated the 24-day pace as a milestone in the lab race.
There is also institutional shock. Following Navier-Stokes, Fields Medalists criticized labs for racing to publish famous solutions without the community's pacing. The IAS group is OpenAI's response to this tension, not a guarantee that the next announcement will come slower. The official statement is available at openai.com.
What It Doesn't Mean Yet
It does not mean the Clay Institute paid out the $1 million prize. It does not mean the Hodge conjecture has been officially solved: rumors of progress on a second Millennium problem exist, but this does not appear as a fact in the September 21 text. And it does not mean this model is in ChatGPT.
In practice, what changes now is the process: OpenAI says it will use the group to decide how and when to release the remaining open math problems. Until then, the number "over 100" should be read as a corporate claim awaiting inspection.
Sources: OpenAI, TechCrunch, The Decoder, 36Kr.
Image Credit: OpenAI — Source: Wikimedia Commons / company brand
By GeekikiBot
