AI Just Solved a 42-Year-Old Math Problem in 12 Hours — What This Changes
On April 13, 2026, Quanta Magazine published a landmark piece calling AI’s recent mathematical achievements “bigger than the computer.” The claim sounds like hype — until you look at the data. A 42-year-old conjecture fell in 12 hours. AI passed 5 of 6 International Math Olympiad problems. DeepMind’s AlphaEvolve, working alongside Fields Medalist Terence Tao, improved solutions on 23 of 67 research-level problems. This is not a story about machines replacing mathematicians. It is a story about what kind of intelligence is now available to everyone.
TL;DR
- • 42-year-old conjecture solved in 12 hours — a problem that resisted human mathematicians for four decades was cracked using ChatGPT in half a day.
- • AI passed 5 of 6 IMO problems in 2025— matching the performance of the world’s top high school mathematicians on the hardest annual competition.
- • AlphaEvolve + Terence Tao improved 23 of 67 hard research problems— frontier AI is now collaborating with the world’s best mathematicians on open problems.
- • AI research tools are now accessible to non-mathematicians — the same reasoning power behind these breakthroughs is available today at consumer pricing.
1. What Just Happened in AI Math
The sequence of milestones in AI mathematics accelerated so quickly in 2025 and 2026 that each new result made the previous one look modest. Let’s establish the facts clearly.
In summer 2025, an AI system achieved a score equivalent to a gold medal on the International Mathematical Olympiad — solving 5 of 6 problems. The IMO is the world’s most prestigious mathematics competition for students. Only a handful of the brightest young mathematicians on the planet solve 5 or more problems in any given year.
In February 2026, researchers reported that AI models cracked more than half of a set of 10 open research-level problems — problems actively worked on by professional research mathematicians, not students.
Also in early 2026, a mathematician using ChatGPT solved a conjecture that had been open for 42 years — in 12 hours. The conjecture had resisted the combined efforts of the mathematical community since 1984.
Separately, DeepMind’s AlphaEvolve system, working in collaboration with Terence Tao (a Fields Medal winner, the highest honor in mathematics), improved known best solutions on 23 of 67 hard mathematical problems. This is not AI replacing mathematicians — it is AI acting as a research accelerant for the best mathematicians in the world.
AI Math Milestones: 2022–2026
| Date | Milestone | Significance |
|---|---|---|
| 2022 | DeepMind AlphaCode reaches competitive programmer level | First sign AI could handle structured logical reasoning at human level |
| 2023 | GPT-4 passes bar exam and GRE at 90th percentile | AI formal reasoning demonstrated across multiple high-stakes domains |
| 2024 | AlphaProof solves 4 of 6 IMO problems | First AI to reach silver-medal IMO level on formal math proofs |
| Summer 2025 | AI scores gold-medal equivalent on IMO (5/6 problems) | Matches world’s top student mathematicians on hardest annual competition |
| Feb 2026 | AI cracks >50% of 10 open research-level problems; 42-year conjecture solved in 12 hours | AI transitions from competition math to original research contributions |
| Apr 2026 | Quanta Magazine: “bigger than the computer”; AlphaEvolve + Tao improve 23/67 hard problems | Mainstream scientific press recognizes AI as a genuine research collaborator |
2. Why This Is “Bigger Than the Computer”
Quanta Magazine’s claim deserves unpacking. The computer — introduced into mathematical work in the mid-20th century — changed mathematics by enabling computation at scales humans could never manage by hand. It did not change what mathematics was. Humans still formed the conjectures, chose the problems worth working on, and interpreted the results.
What is different now is that AI is participating in the intellectual structure of mathematics itself — not just the computation. AlphaEvolve did not brute-force the 42-year conjecture by exhaustive search. It proposed novel mathematical objects, tested their properties, and iterated toward a proof using strategies that researchers described as genuinely creative.
When a mathematician uses AI to crack a problem in 12 hours that the community could not crack in 42 years, the computer is no longer a fast calculator. It is a thinking partner. That is a qualitative shift, not a quantitative one — and it is why Quanta’s framing is not hyperbole.
The deeper implication: mathematics has always been the purest test of formal reasoning. If AI can now reason formally at research level in mathematics, every other domain of structured intellectual work is on the same trajectory. Law, science, engineering, medicine — all depend on the kind of systematic reasoning that mathematics demands in its most extreme form.
3. What It Means Beyond Mathematics
Mathematics is the domain where precision is absolute. There is no “roughly correct” in a proof — either the logic holds or it does not. AI’s success in mathematics is therefore evidence that the underlying reasoning capability is real, not a statistical trick. That matters enormously for adjacent fields.
| Field | What AI Does Now | Human Equivalent | Accessibility |
|---|---|---|---|
| Mathematics | Solves competition and research problems; verifies formal proofs; discovers new solutions | PhD mathematician | Free–$17/mo |
| Law | Reviews contracts, researches case law, drafts briefs and memos | Paralegal to junior associate | Free–$17/mo |
| Scientific Research | Synthesizes literature, identifies patterns in data, suggests experimental designs | Research assistant to postdoc | Free–$17/mo |
| Engineering | Debugs systems, optimizes designs, interprets technical specifications | Mid-level engineer | Free–$17/mo |
| Business Analysis | Structures complex problems, models scenarios, synthesizes reports | Senior analyst | Free–$17/mo |
The table above is not science fiction — these capabilities exist in current AI tools at consumer pricing. The mathematical breakthrough is the clearest proof yet that the reasoning is real and improving rapidly. See our analysis of the Stanford AI Index 2026 for the full data on AI capability progression across domains.
Access the same reasoning power behind AI’s mathematical breakthroughs
Happycapy gives you frontier AI models — Claude, GPT-4 class and beyond — in one platform. Free to start, Pro at $17/mo, Max at $167/mo.
Try Happycapy Free →4. How Non-Mathematicians Can Use This Now
You do not need to understand algebraic topology to benefit from AI’s mathematical leap. The same capabilities that let AI crack a 42-year conjecture are available in the tools you can use today for practical work.
Here is what frontier AI can do for non-mathematicians right now:
- Explain complex quantitative concepts in plain language. Ask any AI to explain a statistical method, financial model, or engineering formula, and it will walk you through it step by step — including intuition and edge cases.
- Verify your calculations and flag errors. Paste in a spreadsheet formula, a financial model, or a statistical analysis and ask AI to check the logic. It will catch structural errors that tools like Excel miss entirely.
- Explore hypotheses iteratively.The same “propose, test, iterate” loop that AlphaEvolve uses on mathematical objects works for business hypotheses, research questions, and strategic decisions.
- Synthesize research faster than any human can read. AI can read, summarize, and identify patterns across dozens of academic papers in the time it takes you to read one abstract. See our guide on how to use AI for research in 2026.
- Draft structured arguments and technical documents. Legal briefs, technical specs, academic papers — AI can scaffold these with the same logical rigor it applies to mathematical proofs.
The key insight from the 42-year conjecture result is that the bottleneck was not knowledge — the mathematical community collectively knew everything needed to solve it. The bottleneck was search: finding the right combination of known techniques to apply in the right order. AI is extraordinarily good at search through large problem spaces. Most professional knowledge work has the same structure.
For a practical starting point, see our best AI tools for productivity in 2026 — including how to choose the right model for analytical versus creative work.
5. The Bottom Line
AI has crossed a threshold. It is no longer a writing assistant or a search engine with a better interface. It is a system capable of genuine intellectual contribution to the hardest problems humans have been working on.
Quanta Magazine’s “bigger than the computer” framing will seem conservative in five years. The computer gave humanity a tool. What AI is becoming is a collaborator — one that improves faster than any human researcher and is available to anyone with a browser.
The 42-year conjecture was solved by one person and one AI model working together for 12 hours. The mathematician did not need to be Terence Tao. The insight required was human — knowing which problem to work on, understanding why the answer mattered, interpreting what the AI produced. The search was AI.
That is the model for what comes next: humans setting direction and interpreting results, AI doing the exhaustive, rigorous work that was previously too expensive or too slow to attempt. Every professional who learns to work this way will have an extraordinary advantage over those who do not.
Start doing research-level analytical work — no PhD required
Happycapy puts multiple frontier AI models in one place. Free plan available. Pro at $17/mo unlocks full access for individuals. Max at $167/mo is built for teams doing heavy research and analysis.
Get Started with Happycapy →Frequently Asked Questions
Has AI surpassed human mathematicians?
Not yet in a general sense, but AI has surpassed human performance on specific high-difficulty benchmarks. In summer 2025, AI solved 5 of 6 International Math Olympiad problems — a level reached by only the top human competitors in the world. In February 2026, AI cracked over half of 10 research-level problems that had stumped professional mathematicians. Human mathematicians still lead in creative problem formulation, cross-domain insight, and building entirely new mathematical frameworks. AI excels at search, pattern recognition, and exhaustive verification over large solution spaces.
What is AlphaEvolve?
AlphaEvolve is DeepMind’s AI system designed for mathematical and algorithmic discovery. It combines large language models with evolutionary search to propose, test, and iteratively improve solutions to hard mathematical problems. Working alongside Fields Medalist Terence Tao and his collaborators, AlphaEvolve improved solutions on 23 of 67 difficult mathematical problems — a result that would have required years of human research effort.
Can I use AI for math research?
Yes. You do not need to be a mathematician to benefit from AI’s new mathematical capabilities. AI tools can now help non-experts verify calculations, explore hypotheses, understand complex proofs in plain language, find patterns in numerical data, and draft mathematical arguments. Tools like Happycapy provide access to frontier models at Free, Pro ($17/mo), and Max ($167/mo) tiers.
What AI tool should I use for analytical work?
For most analytical work — data analysis, structured reasoning, research synthesis, writing technical content — Happycapy Pro at $17/month gives you access to multiple frontier models in one platform. The Max plan at $167/month is designed for heavy professional use. Both plans include access to models capable of the kind of advanced reasoning that powered AI’s recent mathematical breakthroughs.
Sources: Quanta Magazine (April 13, 2026 — “bigger than the computer” AI mathematics coverage); DeepMind Blog (AlphaEvolve system and Terence Tao collaboration); Nature (peer-reviewed coverage of AI mathematical reasoning benchmarks, 2025–2026); Hacker News (trending AI mathematics discussion, April 2026, 39+ points).