A perfect 42/42 at the IMO: reasoning models now clear the hardest school math without tools

The fact, as reported in the note: a model achieved a perfect score of 42/42 at the 2026 International Mathematical Olympiad, a gold-medal-level result, and it did so without external tools or agent scaffolding. The note does not name the model, so we won't guess.

Context for the number: the IMO gives six problems worth seven points each, and a perfect 42 is rare even among the human teenagers who qualify — most gold medals fall short of it. Two years ago frontier systems needed heavy scaffolding (formal provers, code execution, best-of-many sampling) to earn silver. "No tools, no agents" means the reasoning happened inside the model's own chain of thought.

Our reading: this is a genuine capability milestone and a poor proxy for mathematical research. Olympiad problems are hard but bounded — a clean answer exists, verification is cheap, and training data is abundant. Open research is the opposite: ill-posed, unverifiable at a glance, sometimes wrong for years. The honest claim is that machines now handle competition-grade deductive chains unaided; the dishonest one is that they do mathematics.

Why it still matters: the same unaided long-horizon reasoning is what makes AI useful in protein design, drug candidate triage and diagnostic inference — the path we care about most. Watch for the graded solutions to be published. Until third parties see the actual proofs, treat 42/42 as a strong claim rather than a settled fact.