A mathematician working at Anthropic has posted what could become one of the most consequential AI-assisted mathematics results yet: a proposed construction of a complex structure on the six-dimensional sphere, or S⁶.
Levent Alpöge shared the result on X, describing it as “a beautiful new geometric object” connected to a problem he had long loved. He also acknowledged of the role of AI, writing that “Claude really contains multitudes.”
The claim is fairly significant. The question of whether S⁶ admits a complex structure is a famous open problem in differential and complex geometry, often called the Hopf problem. Mathematicians have known for decades that S⁶ occupies a peculiar position among spheres: S² and S⁶ are the only spheres that can carry an almost complex structure, a structure that locally resembles multiplication by the imaginary number (i). S², of course, is the familiar Riemann sphere and has a genuine complex structure. Whether that almost complex structure on S⁶ can be made integrable—meaning that the sphere genuinely becomes a complex manifold—has remained unresolved.
Alpöge’s proposed construction takes an intricate geometric route. The central object is built from a family of complex two-dimensional tori over a modular curve associated with the triangle group (\Delta(3,4,\infty)). The family is then completed at three special points, with carefully chosen degenerations and monodromy. The resulting compact complex threefold is claimed to be diffeomorphic to S⁶.
The full argument reportedly runs to many pages. Alpöge says Claude helped write down the details, while the first one or two pages contain enough of the underlying data for experts to independently reconstruct and check the main computation. A post circulating in the mathematics community points to a full write-up hosted on Alpöge’s site, while discussion on Reddit quickly drew hundreds of reactions and immediately focused on the difficulty of independently checking such a subtle result.
If the proof survives expert scrutiny, the result would settle a problem that has resisted generations of geometers.
And the AI angle could be almost as interesting as the mathematics itself.
A problem with a long history of false starts
The S⁶ problem comes with an important warning label. Claims in both directions have appeared before.
Over the years, mathematicians have published arguments claiming that S⁶ does admit a complex structure, while others have claimed to prove that such a structure is impossible. Several of these arguments later ran into serious objections or failed to achieve broad acceptance. A historical survey of the problem explicitly documents how earlier claimed resolutions failed to produce consensus among experts.
That history makes immediate declarations of victory premature.
At the moment, this is best understood as a proposed solution to a major open problem, rather than an established theorem. The screenshots and the longer PDF are an invitation to the mathematical community to inspect the construction line by line. The real test will come from specialists in complex geometry, algebraic topology and related areas who can verify the argument independently.
This is also why the form in which Alpöge released the result is interesting. Rather than simply posting a claim, he appears to have provided concrete matrices, monodromy data and topological computations that can, in principle, be checked directly. The construction is complicated, but parts of the verification reduce to finite calculations involving explicitly specified matrices and homology groups.
That kind of “certificate” is becoming increasingly important as AI systems start producing mathematics faster than humans can comfortably review it.
Where Claude enters the picture
Alpöge’s description suggests a workflow that is beginning to look increasingly familiar in advanced mathematics.
A human mathematician has a difficult problem, explores ideas, brings mathematical taste and domain knowledge to the search, and then works with an AI system capable of sustaining lengthy symbolic reasoning, checking cases, exploring constructions and writing extensive technical exposition.
In this case, the collaboration apparently led to a construction involving triangle groups, universal families of tori, singular fibres, monodromy and a topological calculation designed to show that the resulting manifold has the same fundamental group and homology as S⁶. Since there are no exotic six-spheres, those topological properties are central to identifying the resulting smooth manifold with the ordinary six-sphere.
The final construction is therefore not simply a clever equation generated from a prompt. It is an elaborate piece of modern geometry, with a proposed object built by gluing together local pieces and then proving that the global topology comes out exactly right.
This is yet another example of AI’s growing role in mathematics.
For years, the benchmark for mathematical AI was the ability to solve Olympiad problems, generate proofs of known theorems or assist with formal verification. Over the last year, however, AI-assisted systems have begun showing up around genuine research problems.
A mathematician released a GPT-5.6-assisted disproof of the Dinitz-Garg-Goemans conjecture, while OpenAI’s claimed to recently solve ten open mathematical problems in one go. Google DeepMind’s AlphaProof Nexus work on open Erdős problems also illustrates a broader shift toward systems that can search through unfamiliar mathematical territory rather than merely reproduce established techniques.
The six-sphere claim belongs to a different category altogether if it holds. S⁶ has been one of the iconic questions in geometry because the object involved is so simple to state and the underlying mathematics is so deep.