AI Models Achieve 35/42 Points for Gold in International Math Olympiad, Signaling New Era for Mathematical Discovery

In a landmark achievement for artificial intelligence, both OpenAI and Google DeepMind announced in July 2025 that their advanced AI models had achieved gold-medal level performance in the prestigious International Mathematical Olympiad (IMO). This marks the first time large language models have performed on par with the world's top high-school mathematicians, solving complex problems that require deep reasoning and creativity.

OpenAI's experimental reasoning LLM reportedly solved five out of six problems, earning 35 out of a possible 42 points. The company stated its model adhered to the same conditions as human contestants, including two 4.5-hour sessions without internet access or other tools. However, OpenAI's early announcement and initial self-grading drew criticism from some within the IMO community, as the company did not officially enter the competition.

Google DeepMind's advanced version of its Gemini model, dubbed "Deep Think," also achieved a gold-medal score of 35 points by solving five problems. Unlike OpenAI, DeepMind collaborated directly with the IMO organization, and its results were officially graded and certified by IMO coordinators. Deep Think utilizes developments allowing it to handle multiple chains of thought in parallel, enhancing its mathematical argumentation capabilities.

The breakthrough has ignited discussions about the future of mathematical research and discovery. As one observer, Haider., noted in a recent tweet, they are "more interested in the release of OpenAI and Google internal IMO gold-winning models than in Gemini 3 or Grok 5." Haider. further emphasized the potential for "novel insights with GPT-5 Pro," suggesting that with the IMO model, "such discoveries will likely become more frequent."

Researchers from both companies believe these AI systems could soon assist mathematicians in tackling unsolved research problems at the frontier of the field. While OpenAI's IMO model is an experimental research system and not immediately slated for public release, both firms plan to offer these advanced tools for testing by mathematicians in the coming months, heralding a new era of AI-human collaboration in scientific exploration.