0
0
On August 1, an AI model did something that made a lot of very smart people stop scrolling. It solved ten math problems that had sat unsolved for years — some of them for decades. Problems that trained mathematicians, the kind of people who spend entire careers on a single question, had tried and failed to crack.
The whole thing cost about $2,000.
That number is almost funny next to what it bought. Not a product. Not a chatbot answer. Ten new pieces of mathematical truth, added to what humanity knows, for less than the price of a used car.
What Actually Happened on August 1
The model is called Astra, an internal, not-yet-public version of OpenAI’s next-generation system. Researchers pointed it at ten open problems spread across eight different fields of mathematics and theoretical computer science — the kind of problems that show up on a “still unsolved” list, not a homework assignment. Astra didn’t just spit out answers. It produced full proofs, and it wrote them in a language called Lean, which lets a computer check the logic step by step and confirm there are no gaps, no hand-waving, no “trust me.”
Then the team published everything on GitHub, in the open, for anyone to verify.
That detail matters more than it sounds like it should. For centuries, a mathematical proof only became “true” once enough other mathematicians read it, argued with it, and eventually agreed it held up. That process can take years. A machine-checkable Lean proof skips the years. Anyone with the right software can run the proof today and get a yes-or-no answer: does this actually work. No committee required.
If you’ve ever wondered what people mean when they say AI is entering a new phase, this is a decent example. Astra wasn’t summarizing something a human already knew, or remixing text it had read somewhere. It was doing what researchers call original work — chasing down an answer that did not exist anywhere in its training data, because no one had found it yet.
The Problems No One Could Crack
It helps to know what “open problem” actually means here, because it’s not the same as a hard exam question. An exam question already has a known answer sitting in an answer key somewhere. An open problem doesn’t. Nobody has the answer. It might not even have one. Mathematicians have been staring at some of these for so long that they’ve become almost like landmarks — famous not for being solved, but for outlasting everyone who tried.
Picture a locked door that generations of very capable people have stood in front of, tried their best keys on, and walked away from. Not because they weren’t smart enough. Because the right key genuinely didn’t exist yet, or nobody had thought to cut one that shape. Astra didn’t kick the door down. It noticed which key fit — across ten different doors, in a matter of days.
BGodInspired has covered a physicist’s theory that all of math might reduce to a single operation — the idea that underneath all the branches and specialties, math might be one connected structure rather than a pile of separate puzzles. Astra’s results are an odd kind of supporting evidence for that idea. A system built to predict patterns in language turned out to be able to navigate pure mathematical structure well enough to find doors nobody else had found. That’s not proof the theory is right. But it’s the kind of coincidence that makes you tilt your head.
Why This Isn’t Really About Replacing Mathematicians
The instinct, reading a headline like this, is to jump straight to “so are mathematicians obsolete now.” Probably not — at least not the way the headline makes it sound. What’s more interesting is the idea researchers keep circling back to: AI as a collaborator that can explore millions of possible directions at a speed no human team could ever match, then hand the promising ones to a person to make sense of.
Think about what that could mean outside of pure math. The same approach — explore an enormous space of possibilities, verify which ones actually hold up — could eventually apply to finding new materials, new drugs, better climate models, problems in physics and engineering that have been stuck for the same reason these math problems were stuck: nobody had found the key yet.
Scientists have already been surprised more than once this year by how much structure was sitting in plain sight, waiting. Astronomers recently found a giant planet they had been staring directly at for eleven years without seeing it — the data was there the whole time. They just hadn’t looked at it the right way yet. Astra’s math problems have the same shape. The answers weren’t invented on August 1. They were sitting there, true, the entire time anyone was looking for them.
Found, Not Invented
That’s the detail worth sitting with for a second, because it’s easy to skim past: Astra didn’t invent those ten answers. It found them.
Invention and discovery are not the same motion. When someone invents something — a language, a law, a rule for a game — it didn’t exist before they made it up, and it could have gone differently. But nobody invents that two plus two equals four, or that a particular equation balances. Those things were already true. They were true a thousand years ago, before anyone had the tools to check them. They’ll still be true a thousand years from now, whether or not anyone bothers to look again.
People have wondered for a long time why that’s the case — why the universe seems to run on a set of rules nobody wrote, sitting there quietly and correctly, waiting for someone patient enough or clever enough to notice. It’s one of the oldest questions there is: not “who built this,” exactly, but “why does any of it hold together so precisely, before anyone was here to need it to.” Most people who’ve spent real time in math or science eventually bump into that question, whether they were looking for it or not. Something bigger than any one discovery seems to be holding the order in place. An AI just found ten more threads of it this week, and it will not be the last time.
If that kind of question — the sense that there’s more order and meaning under the surface of things than we usually notice — is one you find yourself circling back to, BGodInspired put together a short, free “How Close Am I to God?” reflection that a lot of people have found unexpectedly clarifying.
What Happens Next
Nobody fully knows yet, and the people closest to this research are the first to say so. Ten solved problems in one week is not a straight line to solving everything — deep problems tend to get harder to find the further in you go, the same way a spacecraft can take the long way to its actual destination on purpose and still be exactly on schedule. But something did shift this week. A tool built to predict the next word in a sentence turned out to be able to find truths that had outlasted some of the sharpest minds in the world.
Whatever comes of it, ten locked doors just opened. And whoever — or whatever — opened them, the doors were always real. The keys were just waiting to be cut.
Discussion Question
If an AI can find a mathematical truth that existed before anyone discovered it, does that change how you think about where “truth” comes from in the first place — or does it not really change anything for you? Drop your take in the comments.
Share This
- An AI just solved 10 math problems that stumped human experts for decades. For about $2,000. It didn’t invent the answers — it found them. There’s a difference, and it’s worth thinking about. https://bgodinspired.com/index.php/bible-resources/bible-and-science/ai-solved-math-problems-found-not-invented/
- Math truths were true before anyone found them. An AI just found ten more of them in a single week, for the price of a used car. What does that tell us about how much order is quietly sitting there, waiting? https://bgodinspired.com/index.php/bible-resources/bible-and-science/ai-solved-math-problems-found-not-invented/
- “It didn’t invent the answers. It found them.” An AI just cracked 10 open math problems that had stumped human experts for decades — and the distinction between finding and inventing is the whole story. https://bgodinspired.com/index.php/bible-resources/bible-and-science/ai-solved-math-problems-found-not-invented/
Questions People Are Asking
What is OpenAI’s Astra model?
Astra is an internal, not-yet-publicly-released model from OpenAI. On August 1, 2026, researchers announced that Astra had solved ten previously unsolved (“open”) problems across eight fields of mathematics and theoretical computer science, producing formal proofs that were published for public verification.
How did OpenAI verify that Astra’s math proofs were actually correct?
Astra wrote its proofs in Lean, a formal proof language that lets a computer check every logical step for gaps or errors. Because the proofs are machine-checkable, anyone can independently verify them rather than relying solely on human peer review, and OpenAI published the proofs on GitHub for exactly that reason.
How much did it cost to solve these math problems with AI?
OpenAI reported that solving all ten problems cost roughly $2,000 in computing resources — a small fraction of the time and funding that traditional mathematical research on open problems typically requires.
Does this mean AI will replace mathematicians?
Most researchers close to this work describe it differently: as AI becoming a collaborator that can explore an enormous number of possible approaches quickly, with people still needed to direct the search, interpret results, and apply them. The comparison many are drawing is to a research partner that works at a speed no human team could match, not a replacement for human mathematical judgment.
What is the difference between an AI inventing an answer and discovering one?
Inventing describes creating something that didn’t exist before and could have turned out differently, like a rule or a language. Discovering describes finding something that was already true, independent of anyone knowing it yet — the way a mathematical relationship holds regardless of whether anyone has proven it. Astra’s results are discoveries in that sense: the ten answers were true before August 1, 2026; they were simply unknown until then.
