OpenAI says it has discovered a solution to the Navier-Stokes problem, a Millennium Prize Problem that has remained unsolved for roughly 90 years, using an internal AI model paired with 10,000 concurrent agents2.
The company announced the claim in a blog post on Tuesday, stating that the solution was produced by an internal model described as more powerful than the newly released GPT-6 Astra. OpenAI said it began training the internal model on August 28th. The Navier-Stokes problem concerns the mathematical behavior of liquid and gas flow and is one of seven Millennium Prize Problems, each carrying a $1 million reward.
The claim has drawn immediate attention and controversy. The Verge characterized the situation as "drama" surrounding the announcement, and Business Insider framed the development as a "breakthrough drama" warranting broad public attention3. Scientific American coverage, circulated on Reddit, described the potential result as one that could reshape the field1,4.
The excerpt from The Verge's report notes that the model "exhibited unpreced—" before the accessible text cuts off, suggesting OpenAI made further claims about the model's capabilities that are not fully available in the accessible portion of the source.
ANALYSIS The use of 10,000 concurrent agents alongside a model described as exceeding GPT-6 Astra represents a resource-intensive approach that, if the mathematical result withstands peer review, would mark the first time an AI system produced a verified solution to a Millennium Prize Problem. The word "drama" in multiple outlets' framing suggests the claim faces scrutiny from the mathematics community, though the accessible excerpts do not detail specific objections.
OpenAI's decision to disclose the internal model's existence, positioned above its flagship commercial release, provides a data point on the gap between its deployed and research-frontier systems.
The solution has not yet been verified by the Clay Mathematics Institute, which administers the Millennium Prize Problems. Formal verification of Millennium Prize solutions typically requires peer review and a waiting period before any prize is awarded.