OpenAI Says Its Model Answered More Than 300 Open Math Problems. It Posted Them in One Night, and the Math World Is Awed and Angry
A month ago OpenAI said its model cracked one famous problem, and mathematicians revolted. On Tuesday night it released hundreds more, and the fight is now over who checks the work and who sets the rules.
By Sean Anees Saifi · Capital Wealth · Published Sunday, October 11, 2026 · Source: The Wall Street Journal, October 10–11, 2026 weekend edition, whose market figures are the Friday, October 9 close (Exchange)
Key Points
On Tuesday night OpenAI released what it says are AI answers to more than 300 open problems in number theory, algebraic geometry and other fields. The batch ran to roughly 720 papers.
OpenAI says the model tried roughly 4,000 problems, and it published 372 groups of results with summaries of the model's reasoning. It claims advances related to three of the five remaining Millennium Prize Problems, including the Riemann hypothesis. The column doesn't report an independent check.
A Rutgers math professor wrote on X that a human who did one of these proofs would get the Fields Medal on the spot. Many mathematicians reported awe this week. Some rushed their own AI-assisted work out first.
More than two dozen Fields Medal winners signed a furious open letter. OpenAI then helped set up an advisory panel of top mathematicians, and says it followed the panel's publishing guidelines with one exception. It says it'll keep testing hard math on its private models.
The career lesson applies to teachers as much as researchers. When software can do the showpiece task, the scarce skills are checking the work, choosing the question and setting the rules.
300+
open math problems OpenAI says its model answered, released Tuesday night
720
papers in the batch, roughly
4,000
problems the model attempted, by OpenAI's count, for 372 groups of results
3 of 5
remaining Millennium Prize Problems with related advances claimed
OpenAI released roughly 720 papers Tuesday night, with answers to more than 300 open problems and, by its own count, 372 groups of results from about 4,000 attempts.
In one line: OpenAI says its model answered more than 300 open math problems in one night, and the people who spent careers on them must now decide what their job is.
Picture spending your whole career on one hard math problem. Then one night a company posts an answer to it, and to hundreds more. That's what Tuesday night felt like for a lot of mathematicians. A month ago OpenAI said one of its internal models had cracked a Millennium Prize Problem, one of the hardest open questions in math. Mathematicians revolted. Then OpenAI released what it says are answers to more than 300 open problems in a single batch. Ben Cohen tells the story in his Science of Success column.
The batch ran to roughly 720 papers. The problems came from number theory, algebraic geometry and other fields with scary names. OpenAI says its model tried roughly 4,000 problems, and it published 372 groups of results with summaries of the model's reasoning. It claims advances related to three of the five remaining Millennium Prize Problems, including the Riemann hypothesis. One note of caution from us. The column reports the release and the reaction. It doesn't say outside experts have checked the 300-plus results yet. So we write that OpenAI says, and we'd keep it that way until the checkers finish.
The reaction split two ways. Alex Kontorovich, a Rutgers math professor, wrote on X about one of the proofs. A human who did it, he said, would win the Fields Medal on the spot. Many mathematicians said they felt awe this week. Others felt caught in the crossfire of a race between giant companies. More than two dozen Fields Medal winners had already signed an angry open letter. They said the race to solve problems threatens the point of math itself. As rumors of the release spread, some researchers rushed their own AI-assisted work into print before a machine got there first. OpenAI helped set up an advisory panel of top mathematicians. Last week the panel wrote guidelines for how labs should publish. OpenAI says it followed them, with one exception.
The exception is the big one. “We ask them to stop testing advanced mathematical problems on proprietary models,” the mathematicians wrote. OpenAI said it needs to keep testing its internal models on math and science, because that's how it builds the tools faster. The people who do math for a living asked for a rule. The company said no.
Our read
We file this under Career, with a Behavioral lesson too. Most of our readers aren't mathematicians. The lesson still lands. When a tool can do the showpiece task, the scarce skills change. Checking the work becomes the job. So does choosing the question, and setting the rules for how the tool gets used. That's exactly what the mathematicians are fighting about. They didn't ask for the machine to be switched off. They asked for a say in how it's tested and how its results get published. Any profession watching software take a swing at its hardest problem should want the same three things.
For a household, the move is practical. Learn to check the machine's work in your own field, because that's the skill getting scarce. Keep the career plan current the way you keep the roof current. You patch it in dry weather. If your work is changing under you, that belongs in the next review, right next to the retirement date.
What It Means For Your Portfolio
Hold. Learn to check the machine's work in your field
No portfolio move. When software can do the showpiece task, the scarce skills are checking the work, choosing the question and setting the rules, in any field.
General planning principles, not advice for anyone in particular. If AI tools are reaching your profession, spend this year getting good at checking their output. That skill transfers, and it's often what employers pay for once the tool is cheap.
Revisit the career plan and the retirement date together. A field that changes fast can shorten a career or lengthen it. Either way the plan should say which one you're planning for.
