Unsolved Math Problems AI Could Solve Next in 2026

Quick answer: The unsolved math problems AI is best positioned to tackle next include Erdős problems, small Ramsey numbers, the chromatic number of the plane and more cases of the lonely runner conjecture: problems with concrete constructions, computable bounds or proofs software can check. AI has already helped produce major 2026 results, including a Lean-certified Riemann Hypothesis proof (November 21). Formal certification, expert acceptance and official prizes are separate milestones, and no one can reliably predict which famous problem falls next.

Facts checked November 22, 2026. This is a fast-moving story; dates and statuses are given exactly as of this check.

Two years ago, asking which famous math problem AI would solve next sounded like science fiction. In 2026 it is a serious question. AI systems have produced counterexamples to old conjectures, machine-checked proofs and, most dramatically, results on Millennium Prize Problems. This guide gives a clear scoreboard of what is settled, what is claimed and what remains, and explains which problems look most within reach.

The Millennium Prize Problems Scoreboard

ProblemStatus, November 22, 2026
Poincaré conjectureSolved by Grigori Perelman (2003); the only problem Clay lists as solved
Riemann HypothesisFormally certified, official status pending: Levent Alpöge released a Claude-assisted Lean proof on November 20, 2026, and the Lean and Mathlib communities certified the statement and proof on November 21. Clay still lists it as unsolved while broader assessment continues
Navier–StokesOpenAI’s September 2026 blow-up result with an external force; Clay says the problem appears settled and lists it as active, with no prize announced
Yang–Mills and the mass gapOpenAI reported an existence proof for pure Yang–Mills in fall 2026, with a Lean formalization reported in October; the mass gap remains open
P vs NPOpen (2026 claims target a stronger hypothesis, not P vs NP itself)
Hodge conjectureOpen
Birch and Swinnerton-DyerOpen in general; proved for analytic rank 0 and 1

Under Clay’s rules, a solution must be published in a qualifying outlet, at least two years must pass after publication, and it must have general acceptance in the mathematical community. So formal verification and prize eligibility are separate questions, and prize decisions lag far behind announcements. See the Millennium Prize Problems explained.

Which Unsolved Problems Has AI Already Helped Solve?

  • Erdős’s unit distance conjecture (May 2026): an OpenAI model produced a construction contradicting Erdős’s predicted growth rate; mathematicians published a human-verified explanation.
  • The Jacobian conjecture in three variables (July 2026): a counterexample credited to Anthropic’s Claude, announced by Levent Alpöge, with independent machine verification.
  • The cycle double cover conjecture (July 2026): a proof by OpenAI’s model, with a human exposition by Jim Geelen and colleagues.
  • The Riemann Hypothesis (November 2026): a Lean proof certified by the Lean/Mathlib community, with wider review ongoing.

These show three things AI does well: generating constructions, searching for counterexamples and producing formally checkable proofs. See AI math breakthroughs of 2026.

What Makes a Problem “AI-Ready”?

FeatureWhy it helps AI2026 example
A single counterexample would settle itAI is very good at searching huge spaces of constructionsErdős’s unit distance conjecture; the Jacobian conjecture in three variables
It can be stated precisely in LeanA proof checker gives instant, reliable feedback on every stepThe Riemann Hypothesis proof was checked in Lean
The needed theory is already formalizedMathlib, Lean’s library, supplies definitions and lemmasAnalytic number theory, used for zeta-function results
Progress can be measuredA bound that improves gives a clear score to optimizeRamsey numbers, packing problems
It needs new concepts, not just new argumentsHard for AI: there is no clear target to search forHodge conjecture

The Most Likely Next Targets

CandidateWhy AI may helpMain obstacleOur outlook
Erdős problemsConcrete, many small targetsSome need deep theoryStrong, ongoing
Chromatic number of the planeFinite graphs and Lean checksExact value (6 or 7) needs new ideasStrong for bounds
Small Ramsey numbersMeasurable bounds, huge searchesSearch spaces explodeGood for bounds
Lonely runner, more casesEach case is well definedGeneral case needs a new ideaGood for cases
Yang–Mills mass gapBuilds on the 2026 existence resultDeep quantum field theorySignificant but uncertain
Hodge, BSDPrecise statementsTheory only partly formalizedLong-term
P vs NPFormal definitionsKnown proof barriersLong-term, unpredictable

These outlooks are our editorial judgments, not probabilities.

1. Erdős Problems

Paul Erdős left more than a thousand problems, many short and concrete. AI systems solved several in 2026, and the public tracker at erdosproblems.com makes progress easy to verify. Expect more to fall. See Erdős problems solved by AI.

2. The Chromatic Number of the Plane

How many colors do you need so that no two points exactly 1 unit apart share a color? Since Aubrey de Grey’s 2018 computer-assisted construction, the answer has been known to be 5, 6 or 7. In OpenAI’s October 2026 release of research manuscripts, a 62-page paper claims the plane cannot be colored with 5 colors, which would make the answer 6 or 7. Independent community audits in mid-October 2026 report that its Lean proof replays with standard axioms, which is strong evidence, but formal checking is not the same as journal peer review, and the exact value (6 or 7) stays open. Note that this is a different question from Erdős’s unit distance problem, even though both involve points 1 unit apart.

3. Small Ramsey Numbers

The Ramsey number \(R(5,5)\), the smallest party that guarantees 5 mutual friends or 5 mutual strangers, is known only to lie between 43 and 46. The upper bound of 46 came from Vigleik Angeltveit and Brendan McKay, posted in 2024 and published in the Journal of Graph Theory in 2026, using linear programming and huge computer checks. Problems like this, where better bounds are measurable, suit AI search. See Ramsey numbers explained.

4. More Cases of the Lonely Runner Conjecture

Computer-assisted proofs settled 8, 9 and 10 runners in 2026. Each new case is a well-defined target, though a proof for every number of runners needs a new idea.

5. The Remaining Millennium Problems

P vs NP faces known “barriers” showing that whole families of proof techniques cannot work, so it likely needs a conceptual breakthrough. Hodge and Birch and Swinnerton-Dyer rest on deep algebraic geometry that is only partly formalized in Lean, which slows AI checking. The Yang–Mills mass gap is the natural next step after the 2026 existence result. See P vs NP explained and the Yang–Mills mass gap.

6. The Classics: Twin Primes, Goldbach, Collatz

These are easy to state but have resisted every known method. A counterexample to Collatz or Goldbach would be a sensational AI result, but most mathematicians expect both to be true, which means a proof, not a search, is required. See 10 simple unsolved math problems.

Claims vs Verified Results

A 2026 lesson is that announcements and verified results are different things. Use this checklist when you read a headline:

  1. Is there a precise statement? Ideally the exact theorem, written in Lean.
  2. Has the proof been checked? A Lean proof that compiles against a trusted statement is strong evidence; a press release is not.
  3. Have independent experts examined it? Look for named mathematicians, not only the company that made the claim.
  4. Has it been published? Journal review and prize committees take months or years.

A July 2026 Collatz “disproof” that exploited a bug in a proof checker shows why every step matters.

What AI Cannot Reliably Do Yet

  • Invent the deep new concepts that problems like Hodge seem to need.
  • Guarantee that a formal statement says what mathematicians mean; humans must check the statement itself.
  • Explain why a long machine-found proof works in a way people can learn from.

Common Misconceptions

“AI has solved all the Millennium Problems.” No. Several remain completely open, and official prizes depend on long review.

“If AI finds a proof, humans learn nothing.” Not necessarily. Mathematicians are already writing human explanations of AI-found proofs, though some results remain hard to understand.

“Problems that are easy to state are easy for AI.” Not at all. Collatz is easy to state and still open.

Watch: AI and the Hardest Problems in Math

If you learn better by watching, these videos are worth your time.

Google’s AlphaEvolve is making new discoveries in math… (Fireship). A fast overview of how AI systems search for new mathematical constructions.
Has a Millennium Prize Problem been solved by AI? (Stand-up Maths). Matt Parker looks at the 2026 AI claims and what has actually been verified.

Check Your Understanding

Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.

Answer each question and press Submit to see your result right away.

1. Which Millennium Problem does Clay officially list as solved?
Found a bug? Let us know!
2. What makes a problem especially suited to AI search?
Found a bug? Let us know!
3. What is known about R(5,5)?
Found a bug? Let us know!
4. Why do Clay prizes take years?
Found a bug? Let us know!
5. Which part of the Yang–Mills problem remained open in November 2026?
Found a bug? Let us know!

Frequently Asked Questions

Which unsolved math problems could AI solve next?

The best near-term targets are problems where a counterexample or construction would settle the question, or where progress is measurable, such as Erdős problems, small Ramsey numbers and more cases of the lonely runner conjecture.

Has AI solved a Millennium Prize Problem?

AI contributed to 2026 results on the Riemann Hypothesis, Navier–Stokes and the existence part of Yang–Mills. As of November 22, 2026, Clay lists only the Poincaré conjecture as solved and had announced no new prize.

Which Millennium Problems are still open?

As of November 22, 2026, P vs NP, the Hodge conjecture, the Birch and Swinnerton-Dyer conjecture and the Yang–Mills mass gap remain open.

How can I tell if an AI math claim is real?

Look for a precise statement, a proof checked in a system such as Lean, independent expert review and eventual publication.

Why is P vs NP so hard?

Known barrier results show that whole families of proof techniques cannot resolve it, so most experts think it needs a fundamentally new idea.

Keep Exploring

Sources and Further Reading

Related to This Article

What people say about "Unsolved Math Problems AI Could Solve Next in 2026 - Effortless Math"?

No one replied yet.

Leave a Reply