By Brian French | FloridaTechnologyNews.com | September 28, 2026
Quick Answer
In 2026, AI systems moved beyond math competitions and began producing solutions to open research problems. These include dozens of decades-old Erdős problems, a major result about the zeros of the Riemann zeta function, and a claimed solution to part of the Navier-Stokes Millennium Prize Problem. Some results have been verified by leading mathematicians or checked in formal proof software. The biggest claims are still under review.
Key Takeaways
- AI went from a silver-medal score at the 2024 International Mathematical Olympiad to a perfect score in 2026.
- AI tools have helped move roughly 100 Erdős problems to “solved,” though many of those were found through literature searches of existing work.
- OpenAI claims a solution to the forced version of the Navier-Stokes problem. The Clay Institute has opened a multi-year review and has not certified it.
- Anthropic’s Claude helped raise the proven share of Riemann zeta zeros on the critical line from about 41.6% to about 67.2%.
- Florida universities, including UF and FSU, are building AI math tools and preparing students for this shift.
Timeline: AI’s Math Milestones
| Date | Milestone | Status |
|---|---|---|
| Jul 2024 | DeepMind IMO silver (28/42) | Humans translated problems |
| Jul 2025 | Google and OpenAI IMO gold (35/42) | Google’s certified by IMO |
| Jan 2026 | Erdős #397 and #728 solved | Accepted by Terence Tao |
| May 2026 | Erdős unit-distance conjecture disproved | Not yet peer-reviewed |
| Jul 2026 | RedNote AI perfect IMO score (42/42) | Reported |
| Aug 2026 | OpenAI Astra: 10 open problems | Formal proofs; not peer-reviewed |
| Aug 2026 | Claude: two-thirds of zeta zeros | New human proof followed |
| Sep 2026 | OpenAI Navier-Stokes claim | Clay review underway |
From Competition Math to Research Math
Just two years ago, AI’s best math results were modest. At the 2024 International Mathematical Olympiad, Google DeepMind’s AlphaProof and AlphaGeometry 2 scored 28 points, the equivalent of a silver medal, and human experts had to translate the problems into a formal logic language first.
A year later the scores jumped. At the 2025 Olympiad, an advanced version of Gemini Deep Think solved five of six problems for 35 points, a gold-medal score officially certified by IMO coordinators. OpenAI announced an equivalent result the same week. In 2026, RedNote’s dots-note-3.0 model reportedly solved all six problems for a perfect score of 42.
Olympiad problems, however, already have known answers. The more meaningful test is whether AI can solve problems that no one has solved.
The Erdős Problems
Paul Erdős left behind hundreds of unsolved problems when he died in 1996, ranging from minor puzzles to major questions in number theory. They have become one of the main proving grounds for AI.
The early progress came with caveats. According to a tracking page started by UCLA mathematician Terence Tao, AI tools helped move about 100 Erdős problems into the “solved” column between October 2025 and February 2026, but most of that help was an advanced form of literature search, finding existing solutions no one had connected to the problems.
Genuinely new solutions followed. In January, GPT-5.2 Pro produced a proof of Erdős Problem #397, the Aristotle system formalized it in the proof-checking language Lean, and Tao accepted it. Tao said Problem #728 was solved “more or less autonomously” by ChatGPT after some initial feedback.
Tao has also urged caution. He estimated that only about one to two percent of open Erdős problems are simple enough for current AI tools to solve with little human help. A Google team that tested Gemini on 700 “open” Erdős conjectures reached a similar conclusion: many of the problems its AI resolved had stayed open because they were obscure, not because they were hard.
The results grew larger in the spring. On May 20, 2026, OpenAI announced a solution to one of the best-known Erdős problems, the unit-distance problem. In August, OpenAI published a 249-page manuscript claiming that an unreleased model called Astra had solved 10 open problems, with formal proofs, for about $2,000 in computing costs. Peer review is still pending.
The Riemann Zeta Result
The Riemann Hypothesis states that all of the relevant zeros of the zeta function lie on a single line, called the critical line. It remains unproven after more than 160 years.
On August 10, 2026, Anthropic announced that an unreleased research version of Claude had raised the proven share of zeros on that line from 41.6% to 67.2%. Anthropic also said clearly that this method cannot prove the Riemann Hypothesis itself, which requires 100%.
The result has since held up to scrutiny. Mathematician Youness Lamzouri published a new, simpler proof showing that more than 67.25% of the zeros are simple and on the critical line. Anthropic also published a Lean formalization of the theorem.
The Navier-Stokes Claim
The biggest claim came on September 8, 2026, when OpenAI announced that an AI model may have solved the Navier-Stokes problem, one of the seven Millennium Prize Problems, each worth $1 million.
According to OpenAI, roughly 10,000 AI agents working together produced a proof in 88 hours showing that the Navier-Stokes equations can “blow up” in finite time. The claim covers a specific version of the problem. Of the four official Clay formulations, OpenAI’s proof addresses options C and D, which involve smooth external forcing. Options A and B, the unforced versions, remain unresolved.
Verification will take years. The Clay Mathematics Institute said the problem “could have been settled” and started a formal, multi-year review. Clay’s president, Martin Bridson, said the review would be “deliberately unhurried” and “absolutely rigorous.” Credit is also disputed. An NYU mathematician whose related work came first has said OpenAI pursued that approach only after learning about it.
On September 21, OpenAI said the same internal model had resolved more than 100 long-standing open problems. The company has not independently verified that claim or released a problem list with proofs.
Why Florida Should Pay Attention
Weather and fluid modeling. The Navier-Stokes equations describe how fluids move, including air and water. That connects this research to hurricane science, coastal engineering, and aerospace, three fields central to Florida’s economy. A theoretical proof about blowup will not improve next season’s hurricane forecasts, but AI that can reason about fluid equations could eventually speed up the modeling work Florida relies on.
Florida universities are already involved. At the University of Florida, a team led by Professor Vincent Vatter built HaLLMos, an AI tool that gives math students personalized feedback on their proofs without solving the proofs for them. UF also operates HiPerGator, which it describes as the most powerful university-owned AI supercomputer in the country. At Florida State University, mathematician Martin Bauer is organizing a panel on how AI is changing mathematical discovery, training, and careers.
Business applications. Several analysts have noted that if AI can produce research-level proofs for a few thousand dollars, the same approach could be applied to problems like chip design constraints and supply-chain optimization. Florida’s logistics, defense, and space companies all work on these kinds of problems.
How to Judge an AI Math Claim
Readers can check four things:
- Was it formally verified? A proof checked in software such as Lean is much harder to fake than a written argument.
- Was it already solved? Many AI “solutions” turn out to be solutions found in earlier papers.
- Did independent experts confirm it? Acceptance by recognized mathematicians and peer review still matter.
- Is the full problem solved, or a narrower version? The Navier-Stokes claim covers the forced versions only.
What Could Be Next
Mathematicians are already speculating about the remaining Millennium Prize Problems. Many consider the Birch and Swinnerton-Dyer conjecture and the Hodge conjecture the most likely candidates, partly because AI has proven good at finding counterexamples. P versus NP is widely considered the least likely to fall.
Brian’s Take
The key question in all of this is who verifies the result. A system that checks a proof line by line, such as Lean, together with experts who confirm the result is new, is what separates actual progress from a press release. Both of the most credible results in this story, Terence Tao’s accepted Erdős proofs and the zeta-zero bound that an outside mathematician independently re-proved, were confirmed through that kind of review.
The Navier-Stokes claim may hold up, and the Clay Institute’s willingness to open a formal review is significant. The deeper change, though, is economic. When a research-level proof costs about as much as a used car, mathematical capability becomes a commodity that businesses can buy. Florida’s universities, engineering firms, and space companies should start treating AI as a practical tool for technical work, not just a topic in the news.
FAQ
Has AI solved a Millennium Prize Problem?
Not officially. OpenAI claims a proof covering the forced versions of Navier-Stokes. The Clay Institute has opened a multi-year review and has not awarded the prize.
What are the Erdős problems?
Hundreds of open questions left by mathematician Paul Erdős. AI has helped resolve many of them, mostly through literature searches but also through some new proofs.
Did AI prove the Riemann Hypothesis?
No. Claude helped prove that more than two-thirds of the zeta zeros lie on the critical line. The hypothesis requires all of them.
How do mathematicians verify AI proofs?
Through formal proof software like Lean, review by independent experts, and peer-reviewed publication.
Is Florida involved in AI math research?
Yes. UF developed the HaLLMos proof-feedback tool and runs the HiPerGator supercomputer, and FSU is hosting discussions on how AI is changing math research and training.
Sources and Further Reading
- Google DeepMind — Gemini Deep Think IMO gold: https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/
- SCMP — RedNote AI perfect IMO score: https://www.scmp.com/tech/article/3361482/worlds-first-ai-model-earn-perfect-score-maths-olympiad-comes-chinas-rednote
- Scientific American — AI uncovers solutions to Erdős problems: https://www.scientificamerican.com/article/ai-uncovers-solutions-to-erdos-problems-moving-closer-to-transforming-math/
- The Decoder — Terence Tao on GPT-5.2 Pro and Erdős #728: https://the-decoder.com/terence-tao-says-gpt-5-2-pro-cracked-an-erdos-problem-but-warns-the-win-says-more-about-speed-than-difficulty/
- eWeek — GPT-5.2 solves Erdős #397: https://www.eweek.com/news/gpt-5-2-just-solved-a-30-year-math-problem/
- arXiv — Semi-Autonomous Mathematics Discovery with Gemini: https://arxiv.org/html/2601.22401v3
- Quanta Magazine — Why the Erdős problems are falling to AI: https://www.quantamagazine.org/why-the-legendary-erdos-problems-are-falling-to-ai-20260803/
- Tech Insider — OpenAI Astra solves 10 open problems: https://tech-insider.org/openai-astra-solves-10-open-math-problems-2026/
- Yage.ai — Anthropic zeta-zero result analysis: https://yage.ai/share/anthropic-riemann-zeta-search-en-20260813.html
- arXiv — Lamzouri, new proof of the two-thirds result: https://arxiv.org/pdf/2609.02882
- Scientific American — Which million-dollar math problem could AI solve next: https://www.scientificamerican.com/article/which-million-dollar-math-problem-could-ai-solve-next/
- Implicator.ai — Clay Institute on the Navier-Stokes claim: https://www.implicator.ai/clay-institute-navier-stokes-openai-proof-claim/
- DataCamp — Did AI solve Navier-Stokes: https://www.datacamp.com/blog/openai-navier-stokes-math-problem
- Neowin — Clay begins verification process: https://www.neowin.net/news/openai-agent-swarm-triggers-verification-for-navier-stokes-math-problem/
- Tech Insider — OpenAI’s 100+ problems claim: https://tech-insider.org/openai-100-math-problems-solved-24-days-2026/
- University of Florida — HaLLMos AI proof tool: https://news.ufl.edu/2025/12/ai-powered-tool-helps-find-creative-solutions/
- UF — AI research and HiPerGator: https://ai.ufl.edu/research/
- Florida State University — Mathematics at a Turning Point: https://www.math.fsu.edu/~bauer/AI-Math/turning-point-2026.html
About the Author
