A longstanding grand challenge in AI is:
66th IMO held in Sunshine Coast, Australia, July 10 to 20.
For the first time in IMO history, artificial intelligence was formally evaluated alongside human contestants and the results were startling.
submitted full solutions to the 6 official problems, graded independently by former IMO medalists.
Each model scored 35/42, matching the gold medal threshold and outperforming over 90% of human participants.
ref: 2025-international-mathematical-olympiad-results
it’s a strong proof of concept that general reasoning in AI is improving by leaps and bounds:
Sam Altman, CEO of OpenAI, in a public comment on X (Twitter)
underscored the point: “we achieved gold medal level on IMO with a
general-purpose reasoning system… this is an LLM doing math… part of our main
push towards general intelligence.”
AI’s ability to handle language-based logical reasoning has crossed a
critical threshold,
suggesting
machine learning can acquire very deep intellectual skills, not
just pattern matching.
Noam Brown noted that progress has been “fast in math” – in just a couple of years AI models went from grade-school math to saturating high-school benchmarks to now olympiad-level . This pace surprised even researchers: “this result is brand new… It was a surprise even to many researchers at OpenAI”, Brown said. Such rapid advancement has led experts to update their timelines for AGI-related feats, as a challenge thought to be a decade away fell early. As Sam Altman put it, “when we first started OpenAI, this was a dream but not one that felt very \ [near]”, implying that reaching IMO gold was beyond near-term expectations. Now it’s a reality.
ref: https://intuitionlabs.ai/articles/ai-reasoning-math-olympiad-imo
Proof assistant. LEAN. La matematica in linguaggio informatico.