OpenAI announced on September 8 2026 that its researchers solved the Navier-Stokes equation using an internal AI system more capable than GPT-6 Astra, aided by ten thousand simultaneous processing agents; the claim is contested by allegations that the model may have benefited from user data in Codex sessions.
Short answer: OpenAI announced on September 8 2026 that its researchers solved the Navier-Stokes equation using an internal AI system more capable than GPT-6 Astra, aided by ten thousand simultaneous processing agents; the claim is contested by allegations that the model may have benefited from user data in Codex sessions.
OpenAI announced on Tuesday, September 8, 2026 that its researchers have solved a long-standing mathematical puzzle that has resisted solution for roughly nine decades. The claim appeared in a company blog post and was also highlighted earlier by The New York Times and Wired. The problem in question is the Navier-Stokes equation, which governs how liquids and gases move, and it is one of the seven Millennium Prize Problems, each carrying a $1 million award for a correct solution.
According to the post, the breakthrough was achieved using an internal AI system that OpenAI describes as more capable than its recently released GPT-6 Astra model. The system was paired with ten thousand simultaneous processing agents, a scale the company says was essential for exploring the vast space of possible proofs. OpenAI also revealed that work on this model began on August 28, 2026, and that during internal testing it showed what the company calls unprecedented performance on mathematical benchmarks.
The announcement quickly drew attention not only for the technical claim but also for a controversy that surfaced the day before. On September 7, 2026, New York University mathematics professor Tristan Buckmaster, together with Levent Alpöge of Anthropic, published results on a closely related problem. Buckmaster said he reached out to OpenAI after learning the company had become aware of their work. He then alleged that OpenAI had produced a proof for the Navier-Stokes equation using a line of reasoning that he and Alpöge had been developing with the help of OpenAI’s Codex tool and Anthropic’s Claude model.
In his public statement, Buckmaster questioned whether OpenAI had accessed the Codex sessions where he and Alpöge stored their drafts. He recounted asking if the model had been trained on or had viewed those user-generated notes, and said he was told the model did not look up user data. When he pressed further about training procedures, he reported receiving no clear answer.
OpenAI responded to the allegations in its Tuesday blog post, asserting that no specific user data was consulted to arrive at the solution. The company added that, while it considers it unlikely, it cannot completely exclude the possibility that anonymized information derived from how people use its products may have contributed to model improvements. When contacted for comment, OpenAI directed The Verge to a statement posted on its X account that repeats the same points.
Sebastien Bubeck, a technical staff member at OpenAI, echoed the company’s position in a separate remark. He said his team had not seen any of Buckmaster and Alpöge’s work before it was made public the previous night. He noted that, in retrospect, the proofs produced by the two groups differ substantially, both in approach and in the exact results they establish.
Buckmaster countered this defense in a Mastodon post, arguing that OpenAI’s acknowledgment that it cannot rule out the influence of de-identified usage data amounts to an admission that the company benefited from information that became available after his team’s findings were released. He framed the situation as OpenAI “openly admitting they used training data from a period after we found our result.”
Despite the dispute, OpenAI has said it does not intend to claim the $1 million prize attached to the Navier-Stokes Millennium Problem. The company suggests its motivation lies in advancing scientific understanding rather than collecting the monetary reward.
For developers and researchers who build AI systems, the episode highlights several practical considerations. First, the scale of compute, ten thousand agents working in tandem, underscores how resource-intensive cutting-edge mathematical reasoning can become. Second, the debate over data provenance reminds teams that even when models are said not to ingest user-generated content, indirect effects from broader usage patterns can still raise questions about originality and credit. Finally, the decision to forgo the prize illustrates how some organizations may prioritize openness and collaboration over financial incentives, a stance that could influence how future AI-driven discoveries are shared and attributed.
Readers following these developments should watch for independent verification of the proof, keep an eye on how OpenAI addresses data-use concerns in future model releases, and consider how similar controversies might affect trust in AI-assisted research across disciplines
OpenAI announced the breakthrough on Tuesday, September 8, 2026, in a company blog post that was also highlighted by The New York Times and Wired.
OpenAI used an internal AI system described as more capable than its GPT-6 Astra model, paired with ten thousand simultaneous processing agents to explore the vast space of possible proofs.
NYU professor Tristan Buckmaster alleged OpenAI might have accessed Codex sessions where he and Levent Alpöge stored drafts, and questioned whether the model was trained on or viewed those user-generated notes; OpenAI said no specific user data was consulted but could not rule out anonymized usage influence.
OpenAI stated it does not intend to claim the $1 million prize attached to the Navier-Stokes Millennium Problem, saying its motivation is advancing scientific understanding rather than collecting the monetary reward.
Source: The Verge
On September 8, 2026, OpenAI released ChatGPT Images 2.5 with a new Sketch feature that lets users type @Sketch, draw a rough doodle inside the chat, add a natural-language description, and receive an AI-generated image that blends the sketch with the instructions, allowing inline edits and up to 50 % faster generation.
On September 8, 2026, OpenAI released ChatGPT Images 2.5, adding a Sketch feature that lets users draw a rough doodle in a pop-up canvas via @Sketch, then describe style or details to generate a polished AI image.
OpenAI unveiled ChatGPT Images 2.5 on September 8, 2026, delivering noticeably sharper details, richer textures and more natural lighting while cutting generation latency by up to 50 % versus Images 2.0. It also retains core subjects from reference images more reliably and allows precise edits that keep earlier changes intact.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.