The Day the Robots Won a Million-Dollar Math Prize (Sort Of)

 

Copyright: Sanjay Basu

On the morning of Tuesday, September 8, 2026, a group of mathematicians at a company called OpenAI said something remarkable. They had set 10,000 computer helpers loose on a math problem that people had been stuck on for over a hundred years. And the computers cracked it.

The problem is one of six famous puzzles called the Millennium Prize Problems. Whoever solves one gets a million dollars. Only one of the six had been solved before this. Now the computers may have gotten a second.

Let me tell you what the puzzle is, why it matters, and why grown-up mathematicians are both cheering and squabbling about it.

What was the puzzle?

Imagine you are watching water flow down a river, or smoke curl up from a candle, or air moving over an airplane wing. Way back in the 1800s, two scientists wrote down a set of math rules that describe how all of that moving stuff behaves. We call them the Navier-Stokes equations. Say it out loud: "nav-YAY STOKES." Fancy name, simple idea. They are the rules for how fluids move.

Here is the strange part. For 170 years, nobody could answer one simple question about those rules. The question was this: can the math ever go crazy?

Think about it like this. Pretend you have a tiny, tiny drop of water inside a river. The rules say that drop should always move at a sensible speed. But could the rules ever force that little drop to suddenly speed up faster and faster, until it is moving infinitely fast? That would be like a bicycle that keeps going faster forever and never stops speeding up. In math, that runaway moment is called a "singularity." Or, in plain talk, the math "blows up."

For a long time, most experts guessed the answer was no. The water always stays calm and sensible. Now we think the answer is actually yes. The math can blow up. And that is surprising, because the rules themselves look so simple and polite.

So who solved it, the people or the robots?

This is where it gets fun, and a little messy.

The real heroes of the story are two human mathematicians. One is named Diego Córdoba in Spain, and the other is his former student, Luis Martínez-Zoroa. For years, these two did the puzzle the old-fashioned way, using pencils, paper, and their own brains. They even joked about it. Córdoba likes to say, "I don't use AI. I have Luis."

The two of them figured out a clever trick. They imagined stacking up layers and layers of calm water, one on top of another, forever. When you stack them just right, the whole pile does something wild that a single layer never would. They got very, very close to solving the whole puzzle. But not all the way. One piece was still missing.

That last missing piece is what the AI helpers finished off. OpenAI's 10,000 computer agents worked together for about 88 hours. During that time, they sent each other almost 5 million little messages, like a giant classroom of kids passing notes to solve one problem. It cost the company several million dollars in computer power. Then another AI double-checked the whole answer using a special proof-checking language called Lean, which is a bit like a super-strict teacher who catches every single mistake.

So the honest answer is: the people and the robots did it together. The humans built the ladder. The robots climbed the last step.

Wait, why are the mathematicians arguing?

Here is the part I promised not to smooth over.

The same week OpenAI made its announcement, a different team said they had solved closely related problems too. That team was Tristan Buckmaster at a university in New York, working with Levent Alpöge at a company called Anthropic. Buckmaster shared his results just before midnight, only about twelve hours before OpenAI shared theirs.

Now there is a disagreement about who did what first, and whether one team peeked at the other team's work. Buckmaster even hinted that OpenAI's computers might have seen his team's ideas. Nobody has fully sorted this out yet. And one of the mathematicians admitted that one of his own papers, partly written by AI, was so sloppy he apologized for it in public.

I am not going to pretend I know who is right. Nobody knows yet. That is the truth of it. When something big and new happens this fast, the tidy version of the story usually comes months later, after everyone calms down.

Why should you care?

You might be thinking, so what if some water math blows up? My bathtub seems fine.

You are right. Real water is made of tiny molecules, so it does not actually go infinitely fast. This puzzle does not change your bathtub. What it changes is something bigger.

For hundreds of years, the hardest math was done by human minds alone. This week, machines helped crack a problem that had beaten the smartest people on Earth for over a century. That has never happened at this level before. It is a hint that the way humans do deep thinking may be about to change.

And notice one thing before you go. The machines did not do it alone. They stood on the ladder that two patient humans spent years building by hand. The million-dollar question for the next ten years is not really about water at all. It is this: when people and machines think together, who is leading, and who is following?

That one, nobody has proven yet.

---

Source: Konstantin Kakaes, "AI Has Solved One of Math's $1 Million Millennium Prize Problems," Quanta Magazine, September 8, 2026.