The International Mathematical Olympiad is where the planet’s most gifted teenage mathematicians go to feel ordinary. Six problems, two days, no calculators — and problems so hard that solving even one cleanly can move you to tears. So it landed like a thunderclap when an AI system sat the same paper and came within a single point of a gold medal.
What “silver” actually means here
The system solved four of the six problems and earned a score at the silver-medal threshold — better than the vast majority of human competitors who qualify to even be in the room. One of the geometry problems it cracked in seconds. One of the hardest it ground away at for days.
That detail matters. This wasn’t a chatbot pattern-matching its way to a plausible-looking answer. The system wrote proofs in a formal language — every logical step machine-checked against the rules of mathematics, with no room for the confident-sounding nonsense large language models are notorious for. When it claimed a proof, the proof was verified, not just persuasive.
Why mathematicians both cheered and squirmed
A correct Olympiad proof is creative work. It demands the kind of leap — “what if we draw this line, or reframe the problem this way” — that we file under intuition. Watching a machine produce those leaps, and prove them airtight, scrambles the comfortable story that real mathematics is a uniquely human art.
But there’s a more hopeful reading. A tireless system that only ever asserts what it can formally prove could become the ultimate collaborator: check a human’s hundred-page argument overnight, or chase down the tedious lemmas while the mathematician chases the big idea.
The teenagers still win on elegance and speed. But the gap that everyone assumed would take decades to close just shrank to a single point — and it isn’t growing back.