OpenAI’s experimental language model has reportedly solved International Mathematical Olympiad (IMO) problems at a gold medal level, a significant achievement for AI in reasoning capabilities. Although independent verification is pending, the model scored 35 out of 42 points by solving five of six problems in the IMO 2025 competition under strict conditions, which included no external assistance and answers in natural language. Unlike specialized models like DeepMind’s AlphaGeometry, OpenAI’s model utilizes general-purpose methodologies for problem-solving, incorporating innovative reinforcement learning techniques that enhance computational efficiency. This breakthrough comes in stark contrast to other leading AI models, which performed poorly in recent evaluations, failing to secure even bronze medal scores. OpenAI plans to keep this research-focused model under wraps for now, while future iterations could amplify its potential. The implications of this achievement may pave the way for advancements in AI applications in scientific innovation.
Source link
