OpenAI has spent hundreds of millions on math — and Altman says more is coming

OpenAI has spent in the order of hundreds of millions of dollars on mathematical research — and CEO Sam Altman says the company is sitting on far more undiscovered proofs than it has published. Asked on the Cheeky Pint podcast, released Thursday, whether OpenAI holds more results beyond the hundreds of proofs it has already released, Altman replied: "We do. Of course."

The figure gives the first concrete sense of the resources an AI lab is pouring into pure mathematics. It also confirms what frustrated researchers have long suspected: the public releases are only part of a much larger stockpile.

What did Altman actually say?

Altman framed OpenAI's flood of results as a matter of public entitlement to knowledge. "I think, on net, the world deserves to understand how the universe works," he said during the podcast episode. "But I do have empathy for the method."

That empathy has limits. Altman argued the public's right to new scientific knowledge "outweighs" the disruption caused by OpenAI's releases, even as he conceded the company needs to improve how it communicates its findings. "I feel unequivocal on this point: Humanity deserves fundamental knowledge, and we should not be sitting on a major fundamental science discovery," he said. "The world deserves that."

The comments mark one of the first times the OpenAI chief has directly acknowledged the company's impact on mathematicians. That impact has been substantial. A string of genuinely impressive AI-generated results, rolled out in chaotic and poorly communicated fashion, has left researchers bewildered and struggling to keep up — even amid widespread excitement about AI's potential as a tool for discovery.

The takeaway in plain words: Altman admits the communication has been a problem, but he treats the releases themselves as non-negotiable.

Why are mathematicians frustrated?

Mathematicians speaking to The Verge have repeatedly criticized OpenAI for how it announces its results. Their complaints are specific:

• inadequate attribution of who did what;
• little context explaining why a result matters;
• insufficient verification before release.

Researchers say this has left them doing the painstaking work of checking and interpreting OpenAI's findings — unpaid and uncredited — even as the company presses ahead with more. This month alone, OpenAI has released hundreds of mathematical results, often accompanied by little more than brief writeups that researchers have described as poorly written and hard to follow.

In short, the burden of quality control has shifted from the lab that produced the work to the community that must absorb it.

Is OpenAI now withholding results too?

Here the story takes a turn that undercuts Altman's own principle. OpenAI is simultaneously facing criticism for doing precisely what Altman says it should not: sitting on potentially significant findings.

Mathematicians and scientists speaking to The Verge this week said they are frustrated by what they see as OpenAI's gatekeeping of research, following the announcement of three major claimed breakthroughs in mathematics, physics, and computer science. Details for one of those results — involving Yang-Mills theory (the mathematical framework underlying modern particle physics) — have not been released publicly.

OpenAI says the withholding follows the recommendations of a hand-picked group of experts it consulted on how best to release the work. That practice runs directly counter to the advice of a previous group of mathematicians the company had consulted. OpenAI describes the approach as part of a new strategy for communicating its findings.

The contradiction is stark: researchers cannot get full documentation for results already announced, while hundreds of other proofs arrive with minimal documentation at all.

What was the 'Model Name-Hold My Beer' channel?

Altman offered a glimpse of how the flood began. He acknowledged that OpenAI's pursuit of mathematical problems did "get out of control" for a short period of time — but said that was deliberate and necessary to probe the limits of its AI models.

He described an internal Slack channel called "Model Name-Hold My Beer," where employees "were just throwing that at the hardest problems they could" to see how the models would perform. The name suggests the rigor of the process matched the name: an informal stress test, not a research program with external review.

How much did the Navier-Stokes solution cost?

Altman put a number on at least one line item. He estimated OpenAI has spent in the order of hundreds of millions of dollars tackling mathematical problems, including around $10 million — "maybe a little more" — on its Navier-Stokes solution.

Navier-Stokes is one of mathematics' legendary Millennium Prize problems, a set of seven unsolved questions for which the Clay Mathematics Institute offered $1 million prizes each. OpenAI's $10 million spend on a single problem exceeds the prize value tenfold, a signal of how lopsided the resources have become.

The spending figures offer a glimpse of the vast resources AI labs are now bringing to bear on scientific problems. They also raise concerns among researchers that they may be priced out of their own fields — a mathematician with a laptop cannot compete with a lab that can burn $10 million on one equation.

What happens next?

The message to mathematicians hoping for a breather is blunt: there will not be one. Altman expressed sympathy for the disruption, but his commitments ran entirely to improving communication — not to slowing releases, not to withholding the stockpile, and not to funding the verification work his company's output imposes on others.

The unresolved tension is that OpenAI is now criticized both for releasing too much, too fast, and for releasing too little on Yang-Mills. Whichever way the company calibrates its new communication strategy, one thing stays constant: the pace of discovery is set by OpenAI's compute budget, and the pace of verification is left to everyone else.