- In a historic achievement, AI models developed by OpenAI and Google DeepMind have matched gold medal-level scores at the 2025 International Mathematical Olympiad (IMO)—a globally prestigious competition for high school students.
- This marks the first time AI systems have performed on par with top human contestants under official Olympiad-style exam conditions.
About the International Mathematical Olympiad (IMO):
- Established in 1959, the IMO is an annual global math competition.
- Contestants face two 4.5-hour sessions, solving six problems in areas like algebra, combinatorics, geometry, and number theory.
- Each problem is worth 7 points, totaling 42 points.
- In 2025, the gold medal threshold was 35 points.
AI’s Achievement:
- Both OpenAI’s model and Google DeepMind’s Gemini Deep Think scored 35/42—achieving gold medal status.
- They correctly solved five out of six problems.
- This was done within the same time constraints as human participants.
- Gemini Deep Think used parallel reasoning, exploring multiple solution paths at once.
- OpenAI’s model followed step-by-step logic and was verified by past IMO medallists (though not yet officially certified by the IMO).
Comparison with Human Performance:
- India's IMO 2025 team won 3 golds, 2 silvers, and 1 bronze.
- One Indian gold medallist scored higher than the AI models, achieving 37 points.
- Students observed that AI excels in pattern recognition and problem-type memory, but lacks creativity and emotional insight—traits critical in solving truly novel problems.
Significance of the Achievement:
- Marks a leap in AI’s ability to handle abstract mathematical reasoning.
- Earlier AI models needed help to convert questions into formal language—now AI can handle natural language input and generate rigorous proofs on its own.
- This progress has applications in fields like cryptography, theoretical physics, and space research, where unsolved math problems are common.
Limitations and Outlook:
- AI still shows inconsistent performance—it may solve complex questions but fail on simple ones.
- Human creativity, intuition, and originality remain irreplaceable.
- AI is expected to support, not replace, mathematicians
- Checking proofs
- Generating problem-solving strategies
- Assisting in Olympiad training, like how AI supports chess players