Overview

• The model earned 35 out of 42 points at the 2025 IMO by solving five of six problems under the same two 4.5-hour exam conditions as human contestants.
• Three former IMO medalists independently graded its natural-language proofs and unanimously validated the gold-level performance.
• OpenAI confirmed it will withhold the experimental model for several months and that GPT-5 will not include its advanced math capabilities.
• The achievement highlights a shift from task-specific solvers toward general-purpose reasoning in AI, contrasting with systems like DeepMind’s AlphaGeometry.
• Critics have raised concerns about the early announcement overshadowing student competitors and urged an official IMO evaluation of the results.