OpenAI's 10,000-Agent Navier-Stokes Proof Puts Worldcoin (WLD) in Focus

OpenAI Bets on Chip Design

OpenAI is pushing its frontier models beyond chatbots into chip design, life sciences and finance, with the company's finance chief arguing that a model carrying a higher up-front cost can still win by finishing a job in fewer attempts. Chief financial officer Sarah Friar made that case at Goldman Sachs's Communacopia + Technology conference in San Francisco on September 7. The reframing matters for enterprise buyers weighing premium models against cheap ones: instead of asking which model has the lowest price per token, OpenAI wants customers to ask which one delivers the lowest cost per completed task.

The company's own hardware work is its proof of concept. Jalapeño, the first custom inference chip built by OpenAI with Broadcom and reserved exclusively for internal use, moved from design to production in just nine months, per an OpenAI post dated August 25. The company credits AI for compressing design, measurement and verification cycles, and claims the silicon delivers 1.5-1.9 times better AI performance per watt than Nvidia's equivalent GB200 and GB300 parts when running GPT-OSS 120B, DeepSeek R1 and Kimi K2.5. Those are OpenAI's own figures — independent researcher SemiAnalysis observed InferenceX runs but could not reproduce the full benchmark suite.

The pricing argument has partial external backing. In one comparison, GPT-5.6 Luna was priced at roughly $0.18 to complete a standard task against $2.01 for Z.ai's GLM-5.3 — although GLM-5.3 scored 45 on an independent intelligence index versus 38 for Luna, meaning OpenAI wins on cost rather than raw capability. Friar added that cutting the price of the low-end Luna model grew its usage roughly tenfold and that Codex has reached 25 million users. The stakes are enormous: PwC estimates global AI infrastructure capex could reach $31.6 trillion by 2050, with annual spending climbing from about $800 billion in 2026 to $1.8 trillion — even as the OECD notes quality-adjusted model prices fell around 80% between January 2024 and April 2026.

10,000 Agents in 88 Hours

On September 9, OpenAI announced in a post on X that its next-generation model — described as substantially more capable than GPT-6 Astra — has produced a solution to the Navier-Stokes equations, one of the seven Millennium Prize Problems. What should have been a landmark for AI-assisted mathematics immediately turned into a research-ethics dispute instead.

a post on Xhttps://x.com/OpenAI/status/2097375276384567642?s=20

At the center are NYU mathematician Tristan Buckmaster and Levent Alpöge, a mathematician at Anthropic. The pair spent the past year privately collaborating on the problem and made heavy use of AI tools, including Claude and Codex, to work through derivations and organize unpublished material. Once word of their progress began circulating, OpenAI committed its own next-generation model to the same question, deploying roughly 10,000 AI agents that completed a large-scale proof within 88 hours.

Buckmaster says that after learning his findings had reached OpenAI, he contacted the company directly; days later, its research team told him the internal model had already finished a massive proof. His core objection is sequencing: OpenAI poured resources into a rarely used technical direction only after learning of the pair's advance. The company's response has drawn the most scrutiny. OpenAI says the team did not directly access either mathematician's user data while working on Navier-Stokes, but added a pointed caveat — it cannot rule out that de-identified data generated through their use of OpenAI products helped improve the models. It even became an authorship fight: Buckmaster says OpenAI discussed having him help organize its result while excluding Alpöge from co-credit, a condition he ties to Alpöge's employer and refused. We laid out the contours of this clash in our 10,000-agent math proof report, which landed as GPT-6 Astra keeps posting standout benchmark scores. Readers tracking the market in real time can follow live spot and futures prices on Bitget.

Worldcoin's AI-Proxy Trade

Both stories trace one arc: OpenAI is simultaneously seizing the compute layer and accelerating research output, and every expansion of Sam Altman's orbit keeps Worldcoin (WLD) — the proof-of-personhood project he co-founded — trading as crypto's most direct OpenAI proxy. The official announcement confirms only the model result; for the Worldcoin ecosystem, the signal is narrative rather than fundamental. Our desk reads the AI complex as institutionalizing — see our report on how OpenAI and Anthropic pursue investment-grade ratings — a backdrop that has historically lifted the broader altcoin market alongside it. WLD slipped 4.0% over the past 24 hours, yet the token still trades on virtually every major crypto exchange, keeping AI-driven DeFAI flows liquid.