OpenAI Just Quietly Cracked an 80-Year-Old Math Problem That’s Confused Everyone Since the 1940s

So here’s what happened.

OpenAI quietly said this week that their models cracked a math problem people have been stuck on since the 1940s.

Wait, didn’t OpenAI claim this before?

Yeah. And they got embarrassed.

Seven months ago, an OpenAI VP posted that GPT-5 solved ten unsolved math problems. Mathematicians immediately called him out. The problems were already solved. GPT-5 just found the papers. Google DeepMind’s boss called it “embarrassing.” OpenAI deleted the post at that time.

So when they made a similar claim this week, everyone held their breath.

Most AI news is chatbots writing emails or making fake videos of cats doing taxes. Fine. Whatever. But this is different. This is an AI figuring out something that actual human mathematicians couldn’t solve for 80 years.

Let that sit for a second.

OpenAI says its new reasoning model just disproved a famous geometry conjecture that’s been stuck since 1946. It’s called the Erdős unit distance problem. Sounds complicated. Here’s what you need to know:

For 80 years, mathematicians believed square grids were the best way to arrange points to maximize unit distances. OpenAI’s model said nah, and built something weirder. It pulled from algebraic number theory – a completely different branch of math – to construct a better solution.

Eighty years. That means someone’s grandparent could have worked on this problem in their twenties, and now their grandkids are seeing it get solved by a machine.

My take

The GPT-5 mess seven months ago was embarrassing. But this feels like OpenAI learned the hard way. No hype. No dramatic claims. Just the result, the proof, and experts backing it up. That’s how you rebuild trust. Not with press conferences. With receipts.