OpenAI recently proclaimed that one of its models, which outperforms GPT-6 Astra, has triumphantly tackled the Navier–Stokes existence and smoothness problem—a notorious challenge among the Millennium Prize Problems. The model illustrated that, when subjected to smooth external forces, singularities can manifest in fluids within a finite timeframe, thereby satisfying both scenarios C and D as delineated in the Clay Mathematics Institute's problem statement. This feat was accomplished by roughly 10,000 Agents operating concurrently for 88 hours, with GPT-6 Astra overseeing the detection of potential vulnerabilities.

Nevertheless, mathematician Tristan Buckmaster from New York University and researcher Levent Alpöge from Anthropic have voiced their skepticism. They revealed that they had been pursuing research in a similar vein and were on the cusp of a major breakthrough, having archived their research drafts in OpenAI Codex. This has led them to speculate that OpenAI's model might have leveraged their data. Furthermore, Buckmaster highlighted that OpenAI had floated two different release strategies and tried to omit Alpöge from the list of authors.

In response, OpenAI refuted claims of having reviewed the work of the two individuals but conceded that it could not entirely dismiss the possibility that de-identified data might have contributed to refining the model. Relevant personnel also offered divergent accounts regarding the specifics of the dispute. It is worth noting that OpenAI's proof specifically addressed the Navier–Stokes problem in the context of external forces, whereas the variant without external forces—a more pressing concern for the mathematical community—remains unresolved.