OpenAI appears to be closing in on another of the hardest problems in mathematics. The company is close to solving the Hodge Conjecture, one of the seven Millennium Prize Problems, according to reports on September 17, and OpenAI employees expect the problem may be resolved in the near future.
The Hodge Conjecture sits in a branch of mathematics where geometry and algebra meet. First formulated in 1950 by the Scottish mathematician William Hodge, it concerns the relationship between certain topological shapes and the algebraic equations that describe them. The problem asks whether a particular class of objects can always be represented by algebraic combinations, a question that has resisted proof for more than seven decades.
The seven Millennium Prize Problems were set out by the Clay Mathematics Institute in 2000, each carrying a $1 million prize for a correct solution. Only one, the Poincare Conjecture, has been definitively solved, by Grigori Perelman in 2003, who declined the prize. A solution to the Hodge Conjecture would be only the second of the seven to fall, and the first to be credibly attributed to a machine.
OpenAI has made mathematics a central demonstration of its capabilities. The company has previously publicized work on other hard problems, including progress toward other Millennium Prize targets, and it has framed its models as tools for mathematicians rather than replacements for them. A near-solution to the Hodge Conjecture would be the most dramatic result yet in that program.
The reports did not specify how close the company is, or whether the work has been verified by outside mathematicians. A claim of a solution to a Millennium Prize Problem would require formal proof and peer review before the prize could be awarded, a process that has taken years in past cases. What the reports describe is internal expectation at OpenAI that a resolution is near, not a published proof.
The mathematical community has reacted to AI-assisted proofs with a mix of interest and skepticism. A proof generated with the help of a model would need to be checked line by line by human mathematicians, and the question of how much credit the machine deserves would itself become a subject of debate. The Clay Institute’s rules do not specify whether a machine can win the prize, leaving the attribution unsettled.
OpenAI has not commented on the reports. The company has generally been cautious about claims of mathematical achievement, and it has relied on collaborations with established mathematicians to validate its results. Any announcement would presumably come with such collaboration attached, in keeping with the pattern of its prior math work.
The significance of the Hodge Conjecture extends beyond the prize. It is a foundational question in algebraic geometry, a field with applications in cryptography, coding theory, and parts of theoretical physics. A proof would ripple through those disciplines, and a machine-assisted proof would raise the deeper question of whether AI has begun to do mathematics rather than merely assist it.
The moment is part of a broader run of AI results in mathematics. Systems from OpenAI and its rivals have demonstrated the ability to find solutions to competition problems and to suggest novel approaches to open questions, and the boundary between assistance and discovery has become harder to draw. The Hodge Conjecture, if it falls, would mark the point where that boundary moved again.
The Hodge Conjecture is a question in algebraic geometry, the field that studies shapes defined by polynomial equations. It asks whether certain objects that appear when one analyzes a shape using calculus, the so-called Hodge classes, can always be built from algebraic sub-shapes. The conjecture asserts that they can, and proving it for the general case has eluded mathematicians since Hodge stated it in 1950.
The Millennium Prize Problems have a particular mystique because of their difficulty and their price. The Clay Institute set aside $1 million for each in 2000, and in the quarter-century since, only the Poincare Conjecture has been resolved, by the Russian mathematician Grigori Perelman, who declined both the prize and the Fields Medal. A second solution would be an event in mathematics regardless of who or what produced it.
OpenAI’s interest in the problems has been consistent. The company has presented its models as research partners for mathematicians and has publicized work aimed at hard open problems, treating a solution to a Millennium Prize Problem as the clearest possible demonstration of the models’ reasoning ability. A near-solution to the Hodge Conjecture would be the culmination of that program, and a test of how the mathematical community receives a proof that a machine helped produce.
For now, the reports amount to a strong expectation inside OpenAI that the problem will yield soon. Whether the proof arrives, survives review, and is attributed to human mathematicians, to a model, or to both, remains to be seen. What is no longer in question is that the hardest problems in mathematics are now within the reach of the tools the AI companies are building.
