Breakthrough Achievement at World’s Top Math Competition
Artificial intelligence took a major step forward as
Google DeepMind’s and
OpenAI’s latest models each earned gold medal-level scores at the 2025 International Mathematical Olympiad (IMO), a premier global contest for pre-college students. This marks the first time AI systems have reached the prestigious gold medal threshold at the IMO, historically dominated by top human competitors from over 100 countries[1][2][3][4][5].
Performance Details
- Both Google DeepMind and OpenAI models solved five out of six Olympiad problems, earning 35 out of 42 possible points[1][2][3][4][5].
- These models took on the same two-day, two-session structure as human participants, facing two 4.5-hour exams with no internet or external tools[2].
- The 2025 IMO took place in Australia and drew nearly 630 contestants. Only about 10% of human participants received gold medals this year[2].
How the AI Models Worked
The achievement stands out because both companies used advanced general-purpose
reasoning models capable of processing mathematical problems directly in natural language — a major shift from previous methods that required translation of mathematical statements into machine-readable formats[1][3][4][5].
- OpenAI reached the milestone with an experimental system that greatly increased “test-time compute” by running many lines of mathematical reasoning in parallel, allowing the model to “think” longer and iterate over possible strategies[1].
- The models produced natural language proofs, matching or exceeding the performances of most top human competitors without manual problem translation[3].
Implications for Mathematics and AI
Experts suggest this result points to a near future when AI becomes a routine collaborator in mathematical research rather than just a problem-solving tool[1][4]. According to Junehyuk Jung, a leading mathematician at Brown University and a visiting researcher at Google DeepMind, “the moment we can solve hard reasoning problems in natural language will enable the potential for collaboration between AI and mathematicians”[1][4].
This breakthrough also underscores the intense competition between
Google DeepMind and
OpenAI to push AI’s capabilities in complex reasoning — a domain central to human intelligence and expertise[3].
A New Benchmark in AI Capabilities
This milestone at the IMO highlights the rapid progress of AI in sophisticated, creative problem-solving. It also signals the beginning of a new era where AI is poised to tackle unsolved mathematical problems and potentially reshape mathematical discovery itself[1][4][5].
