TLDR: OpenAI has announced that an experimental large language model (LLM) achieved a gold medal-level performance at the 2025 International Mathematical Olympiad (IMO), scoring 35 out of 42 points by solving five of six challenging problems under competition conditions. This marks a significant step towards artificial general intelligence (AGI) and highlights the rapid progress of AI in general-purpose reasoning.
In a groundbreaking announcement, OpenAI revealed that its experimental reasoning large language model (LLM) has achieved a gold medal-level performance at the 2025 International Mathematical Olympiad (IMO), the world’s most prestigious mathematics competition. The AI model scored an impressive 35 out of a possible 42 points, successfully solving five of the six highly challenging problems presented during the competition.
The model was evaluated under the same stringent conditions as human contestants, including two 4.5-hour examination sessions, no access to the internet or external tools, and the requirement to produce detailed, natural language proofs based on the official IMO problem statements. This achievement is particularly notable because the problems demand multi-step reasoning, creativity, and rigorous mathematical proof writing, going far beyond mere computational tasks.
OpenAI researcher Alexander Wei announced the result on X, stating, “Our latest experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad.” He further contextualized the achievement by noting the progression of reasoning benchmarks, from GSM8K to MATH benchmark, AIME, and now IMO, emphasizing that IMO problems “demand a new level of sustained creative thinking.”
The solutions provided by the AI were independently graded by three former IMO medalists, with the final scores based on unanimous agreement, confirming the model’s top-tier performance. This success underscores the rapid advancement of AI in general-purpose reasoning, distinguishing it from specialized mathematical systems like Google DeepMind’s AlphaProof or AlphaGeometry, which achieved silver medal levels in previous years but were designed for specific math tasks.
Sam Altman, CEO of OpenAI, described this result as ‘a significant advancement toward artificial general intelligence (AGI)’ and an ‘indicator of the pace of AI technology advancement over the last decade.’ OpenAI emphasized that this achievement was accomplished through a ‘general-purpose’ model, capable of application in various fields, not just mathematics.
Also Read:
- OpenAI’s New ChatGPT Agent Demonstrates Autonomous Capabilities, Navigating Real-World Tasks with Noted Speed Limitations
- New ARC-AGI-3 Benchmark Highlights Persistent Human Edge in Foundational Reasoning
While this experimental model represents a major research breakthrough, OpenAI has clarified that it is separate from the upcoming GPT-5 and is not planned for public release in the near term. OpenAI researcher Jerry Tworek confirmed that the model received ‘very little IMO-specific work,’ relying instead on continued training of general-purpose base models, and that a public release might be possible by the end of the year. Tworek also noted that several major OpenAI announcements this week, including a general AI agent system and a close loss in a heuristic programming contest, originated from the same reinforcement learning system that powered the IMO success.
