Which open math problems has AI actually solved so far?
As of 22 Oct 2026, the most defensible answer is:
AI has definitely discovered new mathematics, and it has solved some open research problems — but only a small number are clearly established as solved, and several headline claims are still being checked.
The clearest cases where AI has actually solved an open problem
Overview
Open problems AI may have solved
Confirmed
The KLS conjectureMicrosoft Research announced a 2026 proof, with independent validation and machine-checked formal verification
Claimed, not settled
The Erdős unit-distance conjecturereported to have produced a proof/disproof; checked by external mathematicians
existence of non-sofic groups
Connes’s rigidity conjecturebeing disproved
Ehrhart’s volume conjecture
some Erdős problems in extremal combinatorics and Ramsey theory
Drawn by consens.io from this answer of Oct 22, 2026. Every date, number and name is taken from it.
1) The Erdős unit-distance conjecture
This is the strongest widely reported example. An OpenAI model was reported to have produced a proof/disproof of the long-standing discrete-geometry conjecture, and the result was said to have been checked by external mathematicians. If that stands up fully, it counts as a real solution because the conjecture was disproved.
2) The KLS conjecture
Microsoft Research announced a 2026 proof of the 31-year-old Kannan–Lovász–Simonovits conjecture, with independent validation and machine-checked formal verification. This is another genuine open research problem that AI-assisted work appears to have resolved.
3) Several results in OpenAI’s 2026 mathematics release
OpenAI also announced a batch of AI-generated results that it says either resolve or substantially advance long-standing problems. Among the ones that appear closest to “solved” are:
existence of non-sofic groups
Connes’s rigidity conjecture being disproved
Ehrhart’s volume conjecture
some Erdős problems in extremal combinatorics and Ramsey theory
These are serious claims, but the degree of independent acceptance varies. Some are likely genuine new theorems; others are better described as promising but still under scrutiny.
What AI has not clearly solved
AI has not yet produced a broadly accepted solution to the famous flagship problems such as:
the Riemann Hypothesis
P vs NP
the full Navier–Stokes Millennium Problem
Goldbach’s conjecture
the twin prime conjecture
There are reports of related or weaker statements being proved, but those are not the same as solving the canonical open problem.
Important distinction
A lot of AI-math news falls into different categories:
new bound / improved construction: real progress, but not a complete solution
example: DeepMind’s FunSearch work on the cap-set problem
candidate proof or company announcement: potentially major, but not settled until checked
formal proof in Lean / machine-checking: strong evidence, but still needs the statement to match the original problem and the result to be independently vetted
solving contest problems: impressive, but not the same as solving research-level open problems
Bottom line
If you ask, “Has AI actually solved open math problems?” the answer is:
Yes — at least a few, including some major ones — but the number of clearly accepted cases is still small, and many of the most dramatic claims are not yet universally settled.
Only when a check moves the answer. No account needed.
Double opt-in · unsubscribe anytime · your address is only used for these updates.
History
Checked 28 times since Jul 25 · next check Oct 29
How strongly the models agree, per check. Amber points mark checks that changed the answer.
The answer changed
The new answer adds the KLS conjecture as a major AI-assisted problem reportedly proved in 2026, supported by a newly surfaced Microsoft Research source. It also names additional specific results from OpenAI’s mathematics release.
The new answer adds documented AI-generated/formally checked resolutions of open Erdős problems, including a public writeup and broader proof-search results. This strengthens the evidence beyond the standing answer’s framing of outright open-problem solutions as mostly recent claims awaiting broad verification.
The consensus shifted from claiming AI has actively resolved major open problems (like Erdős problems, Navier-Stokes, and Fermat's Last Theorem) to a much more cautious stance that AI has only achieved silver-medal Olympiad levels and improved bounds/constructions, while claims of solving famous open problems lack sufficient independent validation.
The new consensus substantially strengthens the conclusion, asserting that AI has settled major open problems, including a Navier–Stokes result, and resolved numerous Erdős and other conjectures. The old consensus treated such claims as tentative, unverified, and not community-accepted, while affirming only material advances and possible machine-checkable results.
Several AI-generated results are already resolved and validated, rather than only a small number being clearly accepted.
Only a small number of AI-generated results have crossed into independently accepted mathematics; many remain under scrutiny.
OpenAI, Gemini
“the number of results that have crossed from company announcement to independently accepted mathematics is still much smaller—and remains actively contested.”
AI has genuinely resolved a meaningful number of long-standing research problems, with multiple results independently validated or formally checked.
DeepSeek, Grok
“AI has genuinely resolved a meaningful number of long-standing open research problems”
How to verify: Check which claimed results have independent mathematical validation and community acceptance, rather than only lab announcements or formalizations.
