OpenAI Claims 10,000 AI Agents Cracked Navier-Stokes in 88 Hours

For AI researchers, the headline isn't just a math proof—it's a trial run of massively parallel agentic problem-solving at 10,000-agent scale. The dispute over Codex-stored research also tests how training and product data boundaries hold up when frontier labs compete head-to-head on open scientific problems.

On 8 September 2026, OpenAI announced that it had cracked the Navier-Stokes existence and smoothness problem, one of the seven Clay Mathematics Institute Millennium Prize Problems, with an internal system that coordinated roughly 10,000 autonomous AI agents to reach a solution in 88 hours. The company said the effort cost millions of dollars and used a system more powerful than its latest GPT-6 Astra model. If independently verified, the result would be one of the most consequential AI-for-science breakthroughs in the field's short history. Within hours, however, the announcement was overshadowed by a dispute over whether OpenAI had benefited from unpublished work by a rival team at Anthropic.

“

The controversy centers on Tristan Buckmaster, a mathematics professor at New York University, who said OpenAI accelerated its work after hearing that he and an Anthropic researcher were poised to announce their own breakthrough on the problem.

The technical claim itself is striking. OpenAI says the 10,000 agents worked simultaneously, suggesting a shift from single-model inference toward massively parallel, agentic problem-solving. The 88-hour window is presented as evidence of the speed advantage of frontier systems over decades of human mathematical effort, and the disclosure that an internal system exceeded GPT-6 Astra implies OpenAI already has more capable systems in deployment than its publicly named model. But the full architecture, proof structure, and verification details remain unpublished, so the claim currently rests on OpenAI's account rather than peer-reviewed mathematics.

The controversy centers on Tristan Buckmaster, a mathematics professor at New York University, who said OpenAI accelerated its work after hearing that he and an Anthropic researcher were poised to announce their own breakthrough on the problem. Buckmaster noted that the pair's work in progress was stored in OpenAI's Codex model, which is used for writing computer programs, potentially making the material visible to OpenAI staff. His public statement was unusually careful: he said he does not know what OpenAI's model did, does not know whether his data was used, and was not accusing anyone of anything. At a press briefing the same day, OpenAI researcher Sebastien Bubeck denied that the company had used the pair's work or accessed material shared with OpenAI's servers. Yet OpenAI's own announcement added a caveat that could not rule out that data from the pair's use of OpenAI products "helped improve our models."

For the AI and mathematics communities, the episode raises immediate questions about provenance and verification. A correct solution to the Navier-Stokes problem would be a landmark event, but Millennium Prize problems require rigorous scrutiny before any claim is accepted. The Clay Mathematics Institute has not yet indicated whether OpenAI has submitted a formal proof. If the claim survives independent review, it would validate the use of agentic AI systems for open mathematical research. If it does not, or if the provenance dispute deepens, the episode could become a cautionary tale about frontier-lab competition, data separation, and the limits of announcement-driven science.