OpenAI presented a 166-page potential solution to the Navier-Stokes problem in early September, yet mathematicians now question whether the company understands its lineage and methods. The dispute centers on the claimed link to Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa, which experts call superficial. For business users of AI, the episode matters because it shows how hard large-scale agent work remains to audit and explain.
Disputed lineage and massive agent run
According to Scientific American, mathematician Gonzalo Cao-Labora of EPFL said the similarities with the work of Córdoba and Martínez-Zoroa are only surface-level. His formulation was direct: when the details are examined, it is a different approach. The potential proof itself is written in Lean, a formal verification language that checks every logical step but does not make the argument readable to humans on its own. The Clay Mathematics Institute has been reviewing the document since its release.
Before publication, NYU mathematician Tristan Buckmaster accused OpenAI of building on work that he and Anthropic researcher Levent Alpöge had done with publicly available models. OpenAI denied the claim and said it never used Buckmaster data. Camillo De Lellis of the Institute for Advanced Study said a definitive answer is probably impossible because language models are black boxes. That limitation leaves the question of intellectual contribution unresolved even while formal checking continues.
In early October, OpenAI released roughly 400 additional results that the mathematics community can barely keep up with. The company published them as a loose collection on GitHub, a format that angered many mathematicians who felt it cheapened their field. The scale of output contrasts with the slow pace of human review by groups of experts and the Clay Mathematics Institute. The result is a backlog where verification capacity, not model output, sets the pace.
What the dispute means for AI deployment
For companies using AI agents, the compute figures illustrate both capability and opacity. OpenAI deployed about 10,000 model instances for roughly 88 hours and 130 billion tokens to develop the potential Navier-Stokes solution. According to OpenAI researcher Noam Brown, nobody actually knows what that swarm contributed to the result. Systematic comparison tests at that scale have not been studied because running them with thousands of agents costs too much.
No one has tested how long a single agent would take to solve the same problem, and it is nearly impossible to tell what concrete advantage 10,000 agents offer over 1,000. Brown puts less than 10 percent of the success down to the multi-agent setup. He said such systems are new and flashy, so they receive a disproportionate share of credit, while what matters more is the base model itself. That model can grind through tasks over long periods and run in parallel.
The marker to watch is the verdict of the Clay Mathematics Institute review and whether OpenAI publishes controlled comparisons of agent counts. A confirmed proof would strengthen the case for long-horizon formal work, while data on 10,000 versus 1,000 agents would clarify procurement choices. Until then, buyers should treat agent scale as an unpriced experiment rather than a proven lever.
