Google DeepMind's latest iteration of its mathematical reasoning system scored in the top 1% on International Mathematical Olympiad (IMO) problems, which are among the hardest mathematics competition problems given to high school students worldwide. What makes this notable is that IMO problems require creative mathematical insight, not just pattern matching or calculation, involving finding non-obvious approaches to novel problems, exactly the kind of reasoning that was considered uniquely human territory even three years ago. The team submitted their approach to NeurIPS 2026, describing a technique that uses AI to generate candidate proof strategies and systematically explore promising directions.
The machine can now do things we thought only mathematical intuition could do, which means either machines have intuition or we never understood what intuition was.