OpenAI’s New Reasoning AI Wins Gold Medal-Level Performance at 2025 International Mathematical Olympiad
In a groundbreaking achievement, OpenAI’s experimental large language model (LLM) has delivered gold medal-level results at the 2025 International Mathematical Olympiad (IMO), solving 5 out of 6 challenging problems under the same constraints as human participants. This marks a significant milestone in AI's ability to perform creative, complex mathematical reasoning akin to top human mathematicians.
What is the IMO Achievement?
The IMO is one of the world’s most prestigious mathematics competitions featuring the toughest problem-solving challenges. OpenAI’s AI scored 35 out of 42 points, exactly meeting the threshold for a gold medal. Conducted in two timed contest sessions without external tools, the AI generated full natural language proofs for each problem, which were verified by three former IMO gold medalists. This is the first time an AI has matched elite human performance on this contest’s problems.
Training and Technology Behind the Model
OpenAI’s reasoning AI is a general-purpose large language model based on the transformer architecture similar to GPT-4 and GPT-5. Unlike previous math-specific tools, this system is not a symbolic math engine but a language-based model trained to predict tokens with reinforcement learning and extensive reasoning data. It can sustain hours-long intensive reasoning, enabling it to explore complex solution strategies and avoid hallucination errors common in LLMs.
How OpenAI’s AI Compares and Implications
Alongside Google DeepMind’s Gemini Deep Think model, which also achieved gold-level scores at the 2025 IMO, OpenAI’s AI demonstrates the rapid progress of AI in symbolic and abstract thinking fields beyond just language understanding. This success is seen as a leap toward artificial general intelligence (AGI), with major potential impacts for education, scientific discovery, and collaborative human-AI mathematical research.
Future Prospects and Expert Opinions
Experts acknowledge this breakthrough as a sign that AI is nearing a role in assisting high-level mathematical research and learning. OpenAI’s research team emphasized that the model’s capability was developed with new general-purpose techniques making it vastly better at verifying complex proofs. Continued advancement is expected, with possibilities for AI to help innovate in science and mathematics, transforming education and research paradigms.
This historic milestone underscores the transformative potential of AI reasoning, not just in solving math problems but also in fostering creative thinking processes akin to the brightest human minds worldwide.