Editorial illustration for AI Fails to Crack Math's "Major Advance" Problems, USD 1M Prizes Remain
AI Cracks 79-Year Math Mystery, $1M Prize Awaits
AI Fails to Crack Math's "Major Advance" Problems, USD 1M Prizes Remain
The Unit Distance Conjecture sat unsolved for 79 years before OpenAI published a counterexample to it in May 2026, cracking a problem tied to Paul Erdos that had outlasted generations of mathematicians. Within a week, human researchers had taken the model's proof technique and used it to knock down a second major conjecture. Since then, results have piled up fast: Epoch AI, the group behind the FrontierMath benchmark, announced a second solution to one of its "Open Problems" test cases, and OpenAI followed with a new model, Astra, introduced alongside ten solved problems of varying difficulty.
None of this has touched the roughly USD 1 million Millennium Prize problems, which remain as unsolved as ever. But the pace of smaller breakthroughs has forced a reckoning inside a field that prizes human ingenuity above almost anything else. Some mathematicians shrug it off as another tool, useful but limited to their corner of research.
Others describe something closer to professional vertigo, watching machines chip away at problems built specifically to resist easy answers. Abhishek Saha, a math professor at Queen Mary University of London, has been among those weighing in publicly on what it means for the discipline.
AI models are cracking conjectures that stumped humans for decades. Reactions in the math community range from shrugs to an existential crisis as the profession grapples with its own possible obsolescence.
Why this matters
The gap between "AI solved a 1946 conjecture" and "AI can touch the Riemann Hypothesis" is the whole story here, and it's easy to blur the two in a headline. What OpenAI's Astra and similar systems are doing, clearing "Solid Progress" or "Advance" tier problems, is genuinely useful pattern-matching over decades of accumulated math literature. That's not nothing.
But the Clay Institute's six remaining Millennium Prize Problems sitting untouched, and zero wins in the "Major Advance" or "Breakthrough" categories, tells us where the real ceiling is right now. For founders pitching "AI that does research mathematics," that distinction should shape the product claims and the funding pitch. For researchers, it's a decent signal of where automated theorem-proving actually adds value today: mining known techniques faster, not generating the kind of conceptual leap that wins a Fields Medal.
Watch the tier breakdown in future benchmark updates, not the headline count of "problems solved," since that number will keep climbing while the categories that matter most stay stuck at zero.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives MiniMax H3 Video Model Generates 2K Clips, Priced at USD 1.95 for 15 Seconds
From Archives Supabase Launches Evals to Benchmark Claude, Codex, and OpenCode on Real Tasks
Common Questions Answered
What was the Unit Distance Conjecture and how did OpenAI solve it?
The Unit Distance Conjecture had remained unsolved for 79 years before OpenAI published a counterexample to it in May 2026, finally cracking a problem tied to mathematician Paul Erdos. This breakthrough was significant enough that human researchers were able to take the model's proof technique and use it to solve a second major conjecture within just one week.
Why haven't AI models solved any of the Clay Institute's Millennium Prize Problems despite recent successes?
While AI systems like OpenAI's Astra have successfully solved problems in the 'Solid Progress' or 'Advance' tier, the six remaining Millennium Prize Problems remain untouched because they represent a significantly higher level of difficulty classified as 'Major Advance' problems. The gap between solving decades-old conjectures and tackling these top-tier problems represents the full scope of mathematical challenge that current AI cannot yet bridge.
What is the FrontierMath benchmark and what role has it played in measuring AI's mathematical abilities?
The FrontierMath benchmark, created by Epoch AI, is a test designed to evaluate AI systems' capabilities on open mathematical problems. The group announced a second solution to one of its 'Open Problems' test cases, demonstrating that the benchmark is actively tracking AI progress in solving previously unsolved mathematical challenges.
How are mathematicians reacting to AI's ability to crack unsolved conjectures?
Reactions in the mathematics community range widely from indifference to existential concern, as the profession grapples with the possibility of its own obsolescence. However, experts note that while AI's pattern-matching capabilities over decades of accumulated mathematical literature are genuinely useful, the technology has not yet reached the level needed to solve the most difficult 'Major Advance' tier problems.
What is the significance of AI solving problems that stumped humans for decades?
AI solving conjectures that remained unsolved for decades represents a genuine advancement in pattern recognition and mathematical problem-solving, as these systems can synthesize and apply techniques from vast amounts of mathematical literature. However, the article emphasizes that this capability should not be conflated with solving the most prestigious unsolved problems, as there remains a substantial gap between clearing mid-tier problems and tackling the Clay Institute's Millennium Prize Problems.
Further Reading
- An OpenAI model has disproved a central conjecture in discrete geometry - OpenAI
- Remarks on the disproof of the unit distance conjecture - arXiv
- OpenAI Model Cracks 80-Year Erdős Conjecture, Verified by Its Harshest Previous Critic - TechTimes
- An AI solution to an 80-year-old problem has shocked mathematicians - The Conversation
- An AI math breakthrough sparks calls for new guardrails - Science News
