Altman ties $10M Navier-Stokes proof to OpenAI's next GPT
What's the deal? OpenAI chief executive officer Sam Altman said the internal model that produced the company's claimed Navier–Stokes proof is the basis for its next GPT, which he signalled could release soon. He made the comments at the ADIPEC conference, days after discussing the proof on Stripe's Cheeky Pint podcast.
What's the backstory? In September, OpenAI said its agents solved a version of the Navier–Stokes problem, one of seven Millennium Prize Problems the Clay Mathematics InstituteDealroom has a profile for this one. Try Dealroom → set in 2000, each carrying a $1 million prize. OpenAI said it would forgo the prize.
How it ran: Roughly 10,000 copies of the internal model worked as coordinated agents, reaching the result about 88 hours after launch. The run exchanged 2.7 million messages and generated some 130 billion output tokens; a separate 17-hour check in Lean verified each logical step by machine.
The cost: Altman said the work ran "order $10 million," or "maybe a little more." He expects to redo it at a hundredth of that within a year, driven by "smarter models and better approaches."
He framed the spend as the price of working without a playbook. "Look, we did let it get out of control for one week on the math. Intentionally so," he said, adding the team understood "absolutely nothing about the scaling laws of teams of agents" when the run began.
What could go wrong? The claimed result is not yet independently settled. A preprint from mathematicians at King's College London and Cambridge found discrepancies between OpenAI's human-readable proof and its Lean-checked version — without establishing the underlying result is wrong. Others have said the long proof is hard to interpret and has yielded little insight so far.
Why it matters: Altman's comment is a preview, not a launch. OpenAI has not given the next GPT a name, release date, price, or public benchmarks.
The signal: Tying the next model to a splashy math result suggests OpenAI may market it around intensive, multi-agent research — even as the headline achievement remains under scrutiny.
Read more: rdworldonline.com
