Vintage: 2026-09. Issuer claim dated September 8, 2026. Primary evidence is OpenAI’s On the Navier–Stokes Millennium Prize Problem plus navier-stokes.pdf abstract (pdftotext first pages) and official X, as synthesized in 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5. Issuer claim / Lean formalization pending independent math review. Forced C/D ≠ Clay unforced A/B.

OpenAI forced Navier–Stokes claim (Clay C/D, smooth force)

One-line summary: On 8 Sep 2026 openai published an issuer claim of an AI-produced solution to Navier–Stokes existence and smoothness framed as Clay/Fefferman C and D: finite-time singularity from rest under a smooth external force, energy finite, Lean formalized. OpenAI says it does not intend to claim the Millennium Prize.

The insight

This is a dated issuer math claim, not an independently reviewed theorem and not the unforced A/B prize statements. Grain: forced NS (smooth force) = C/D with forcing. Do not write “unforced Navier–Stokes solved” or “Clay prize won.” Keep three Euler/NS lanes separate: this page (OpenAI forced NS C/D); openai-unforced-euler (OpenAI unforced Euler side-result); buckmaster-alpoge-forced-blowups (Buckmaster–Alpöge forced Euler/Boussinesq/IPM); anandkumar-unforced-euler-candidate (PINN unforced Euler candidate).

It is a mid-2026+ instance of the verifiable-layer spike on ai-math-capability-jaggedness (Lean check claimed). It is not a definition- or conjecture-generator result and does not close can-llms-choose-the-right-research-question. Credit-dispute narrative lives on openai-fluid-proofs-credit-dispute — do not flatten this page’s math grain into plagiarism-proven or allegations-false.

Evidence

All bullets are from 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5 (method: grok-bot; x_video: false). Permalinks and issuer URLs live on the source page.

Issuer blog + PDF abstract

• From 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5 (On the Navier–Stokes Millennium Prize Problem, dated September 8, 2026; clip curl 2026-09-08): OpenAI claims a solution to Navier–Stokes existence and smoothness (Millennium): initially smooth fluid at rest develops a singularity in finite time; smooth force applied; energy remains finite. Framed as establishing Clay/Fefferman statements “C” and “D” (Clay NS statement PDF).
• From the same source (same issuer page): model is internal, “significantly more capable than GPT‑6 Astra”; training ongoing since ~Aug 28; “step-function” on many internal benchmarks (issuer).
• From the same source (same issuer page): multi-agent eval of Millennium problems after Sep 1 rumors; NS group ~10,000 concurrent agents; ~88 hours to resolution (arrived Sat Sep 5); Lean formalization ~17 hours via GPT‑6 Astra; NS effort ~2.7M messages / ~130B output tokens (all problems ~4.9M / ~300B).
• From the same source (same issuer page): physical picture — inward-spiraling vortex elongated “like spaghetti”; motion terms cancel so external force stays smooth while velocity unbounded.
• From the same source (same issuer page): “We do not intend to claim the Millennium Prize for this result.” Goal framed as reporting model progress / pacing responsibility.
• From the same source (navier-stokes.pdf abstract, first pages, pdftotext): for every viscosity ν>0, construct force in Cc∞ and smooth fields from rest with bounded kinetic energy but limsup ‖u‖∞ = ∞ — establishing alternative (C); compact support also yields (D) on the torus.
• From the same source: Lean/repo github.com/openai/NavierStokesAndEuler. Companion unforced Euler writeup is not this page — see openai-unforced-euler.

Official X (text grain; NS announce is photo, not video)

• From the same source (@OpenAI, 2026-09-08 17:20:56Z): sharing NS Millennium solution; agents + next-gen model ≫ Astra; ~90-year open problem framing. (photo attached — not video)
• From the same source (@OpenAI): step-function model; 88 hours, ~10,000 coordinating agents; frontier eval safeguards.
• From the same source (@OpenAI): analytical proof + Lean; singularity via NS dynamics; vortex/spaghetti; link to PDF.
• From the same source (@OpenAI): understand model / guide and pace capability; steerable/accountable systems; link to issuer blog.

What this source does not establish

• Not Clay unforced A/B. Forced C/D with a smooth external force ≠ “no force” prize statements. Discourse saying “Navier–Stokes solved without forcing” or “Clay prize won” overclaims relative to issuer + Clay statement.
• Not independently reviewed. Lean repo + PDFs are public; community/Clay review status is not established in this clip.
• Not a rewrite of gpt-6-astra. Astra is named for ~17h Lean formalization; the NS model is an unnamed internal model claimed ≫ Astra.
• PDF body beyond the abstract was not re-parsed page-by-page in this clip.
• No ticker.

Contradictions / tensions

• Forced vs unforced vs “Millennium solved.” This page = smooth force (C/D). Keep separate from openai-unforced-euler, buckmaster-alpoge-forced-blowups, and anandkumar-unforced-euler-candidate. Not collapsed.
• Issuer Lean vs independent verification. Public repo is not the same as community/Clay review.
• Prize disclaimer vs discourse “prize won.” OpenAI explicitly not claiming the prize.

Open questions

• Does independent math review confirm the Lean statements match the intended C/D writeup?
• How does this forced-C/D object relate to Buckmaster’s morning report of a ~100-page forced-NS writeup he had not seen?

Related

• openai-unforced-euler
• buckmaster-alpoge-forced-blowups
• anandkumar-unforced-euler-candidate
• openai-fluid-proofs-credit-dispute
• openai
• gpt-6-astra
• openai-research-acceleration
• openai-alien-mind
• ai-math-capability-jaggedness
• mathematics-in-the-age-of-ai
• math-verification-loop-to-jagged-ai-progress
• automated-ai-research-llm-capability-boundary
• can-llms-choose-the-right-research-question — status stays open; do not close as yes
• claude-flt-lean-formalization
• sam-altman
• sebastien-bubeck