News Summary

OpenAI has announced that its new AI language model has achieved gold medal-level performance at the International Mathematical Olympiad (IMO), a standard reached by fewer than 9% of human contestants annually. The announcement, made by OpenAI researcher Alexander Wei, came prematurely, disregarding an embargo by IMO organizers to wait until July 28. The model tackled the competition’s six proof-based problems under conditions similar to human competitors, with 4.5-hour sessions and no internet or calculators. However, the legitimacy of OpenAI’s claims is questioned as it self-graded its results. OpenAI plans to release the proofs and grading rubrics for public scrutiny. This achievement marks a significant shift from previous AI attempts at IMO problems that relied on specialized systems, as OpenAI’s model processed problems in plain text and generated natural-language proofs. The announcement follows Google’s prior achievement with its AlphaProof and AlphaGeometry 2 models, which took longer and needed human assistance. OpenAI noted that while its next major model, GPT-5, is forthcoming, this current version is experimental and not yet consumer-ready.