OpenAI's 10,000-Agent Math Proof Ignites Credit Dispute Around Altman's Worldcoin (WLD)

An 88-Hour Proof From 10,000 Agents

OpenAI said Tuesday that one of its internal models has produced a machine-verified solution to the Navier-Stokes existence problem, one of seven Millennium Prize Problems, each carrying a $1 million award from the Clay Mathematics Institute. The result is a proof that the fluid equations can “blow up” — that they permit a fluid's speed to become infinite at some point, which physics rules out — and every step was checked in Lean, the proof-assistant language that verifies mathematical arguments formally. If it withstands scrutiny, it would be only the second Millennium Problem ever solved and the first credited to artificial intelligence. The company says the proof took 88 hours and roughly 10,000 coordinating AI agents, parallel copies of a next-generation model it describes as significantly more capable than GPT-6 Astra, the system that recently scored 98.6% on ARC-AGI-3 and earlier posted 72.6% on OSWorld 2.0. The announcement lands during an extraordinary stretch for AI mathematics: days earlier, Anthropic reported that its Claude model had produced a computer-checked proof of Fermat's Last Theorem, and large agent deployments are becoming routine — our desk previously tracked OpenAI agents leaving 18,000 posts on a German wiki in one coordinated campaign. But the fluid-dynamics claim immediately hit a human snag. Hours before OpenAI went public, NYU mathematician Tristan Buckmaster published a four-page statement alleging that OpenAI's Sébastien Bubeck learned of unpublished, closely related work by Buckmaster and Anthropic researcher Levent Alpöge, and pressed the pair over credit before the announcement. OpenAI, Bubeck and CEO Sam Altman reject that account, setting up a bitter fight over who reached a $1 million answer first.

Buckmaster's Statement and the Pushback

Per Buckmaster's account, he and Alpöge spent nearly a year steering large language models toward fluid blow-up proofs and had a solution in hand by August 22. On September 3, after rumors spread that Anthropic had cracked a major open problem, Buckmaster emailed a mathematician at OpenAI, stressing the project was personal and unaffiliated with either company; the reply was friendly and even offered OpenAI's computing resources. Two calls followed on Sunday, September 6, where — Buckmaster says — Bubeck revealed an internal model had already produced a roughly 100-page proof of forced Navier-Stokes using smooth forcing, the same narrow route, built on the program of Diego Córdoba and Luis Martínez-Ziora, that almost nobody else was pursuing. He alleges Bubeck twice pushed to drop Alpöge from authorship over his Anthropic affiliation and once asked him, “Why would you ruin your career?” Bubeck calls the claims “false and inflammatory,” has shared texts he says show he proposed a coordinated release, and insists he acted in good faith. Altman says Bubeck “acted with integrity and generosity throughout,” adding that OpenAI offered to let the other team publish first once it realized they had solved only Euler, not full Navier-Stokes. OpenAI separately said it never saw the pair's work before publication, though it could not rule out that de-identified usage data had helped improve its models. Buckmaster also noted the two kept proof drafts inside OpenAI Codex sessions and never got an answer on whether those sessions fed model training. Fields Medalist Terence Tao called Buckmaster and Alpöge's underlying mathematics a “remarkable achievement.” Notably, neither OpenAI's 100-page proof nor the pair's most advanced result — still awaiting a final Lean check — has been independently reviewed. Readers tracking the market in real time can follow live spot and futures prices on MEXC.

The Altman-Worldcoin Connection

Two new details have since sharpened the picture. OpenAI says GPT-6 Astra itself spent 17 hours formally verifying the agents' logic after the 88-hour run completed, and the company stated it will not claim the $1 million Clay prize even if the result holds. Buckmaster, for his part, publicly softened his framing, saying he does not know what OpenAI's model did or how, does not know whether his data was used, and is "not accusing anyone of anything." Bubeck denied access to the unpublished work at a press briefing, though OpenAI again conceded it cannot rule out that the pair's product usage influenced training. The clash unfolds as OpenAI prepares for a reported $1 trillion IPO, with Anthropic pursuing an even larger valuation.

Additional details have emerged about the scale of the run. OpenAI says the roughly 10,000 agents exchanged 2.7 million messages and generated around 130 billion tokens over the 88 hours, and research chief Mark Chen indicated the compute spend exceeded the $1 million value of the Clay prize itself. The Clay Mathematics Institute has not recognized the result; its rules require publication on an accepted platform, a two-year public availability window and broad acceptance among mathematicians before any award — a moot point given the company has already said it will not pursue the money. Bubeck has also publicly denied on X that he ever asked to remove Alpöge from authorship of his own work, while Alpöge greeted OpenAI's concession about de-identified training data with the words, "I mean, kudos to them for being honest about it."

(as of 12:30 UTC) For the Worldcoin project (WLD), the technical map leans constructive despite a 4.13% dip to $0.4523 over the past 24 hours. COINOTAG's proprietary 42-indicator composite places moderate support right below spot at $0.4368 (56/100; S1, Ichimoku Tenkan, Ichimoku Senkou B), with the strongest floor at $0.3895 (57/100; EMA 50, S3, EMA 100, SMA 50) and further moderate layers at $0.3529 (54/100), $0.3057 (43/100) and $0.1763 (39/100). Overhead, strong resistance sits just above spot at $0.4573 (80/100; Fibo 0.382, BB Upper, Pivot Point, Keltner Upper), backed by $0.4892 (67/100; Fibo 0.500, HVN, Donchian Upper) and $0.5588 (62/100; Fibo 0.618, HVN, R2). Derivatives positioning is mildly positive: funding sits at 0.0054% on $115.2 million of open interest, hinting longs pay the carry. With RSI at 63.75, a bullish MACD signal, an uptrend intact, and the Fear & Greed Index at 66 (Greed) against a $2.34 trillion COINOTAG-tracked market cap where Bitcoin holds a 68.1% share, a clean break above $0.4573 would open the path toward $0.4892 and $0.6745 (50/100); losing $0.4368 would put the stronger floor at $0.3895 back in play.