OpenAI claims unreleased model resolved Navier-Stokes Millennium Prize problem; Anthropic-linked mathematician disputes credit and priority
2026-09-08
OpenAI said an unreleased internal model, run as ~10,000 collaborating agents over 88 hours (with Lean verification via GPT-6 Astra over an additional 17 hours), produced a resolution to the Navier-Stokes existence and smoothness Millennium Prize problem, per OpenAI's own account relayed by Simon Willison and Latent Space. NYU mathematician Tristan Buckmaster, who had been working on a related result for nearly a year with Levent Alpöge (an Anthropic employee), told TechCrunch that OpenAI began its effort only after hearing rumors of their unpublished breakthrough, declined to credit Alpöge as a co-author due to Anthropic's competitive status, and gave evasive answers about whether the model had access to their private Codex session data.
Significance 4: A frontier-capability claim of this magnitude would normally rate higher, but per the rubric lab-reported claims cap at 4 until independently reproduced, and the math community has not yet verified the result; the credit/priority dispute is a real, multi-outlet-corroborated strategic-behavior signal in its own right.
Implications · machine-drafted, not owner judgment
If the mathematical result holds up, it meaningfully raises the ceiling on what unreleased frontier models can do on genuinely hard, open-ended reasoning problems — relevant to the owner's hard-problems verdict once such a model ships publicly. Separately, OpenAI's handling of the priority dispute (declining co-authorship to a rival-lab employee, evasive answers on data access) is a live test of research-community trust norms as labs race on the same problems, and feeds directly into the credibility question of whether OpenAI's AGI-adjacent messaging reflects substance or positioning.
- Independent mathematician verification or peer review of the claimed proof
- Publication of the Lean-verified formalization for outside scrutiny
- Any further disclosure on whether de-identified Codex session data influenced training
Sources
- curator OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next retrieved 2026-09-09
- curator On the Navier–Stokes Millennium Prize Problem retrieved 2026-09-09
- press OpenAI fought dirty on career-making math problem, says NYU mathematician retrieved 2026-09-09
- curator OpenAI Does Math, Reward-Hacking, Meta Launches Personal Agent retrieved 2026-09-09
Affected axes
Hard reasoning