AI Scores Gold at Math Olympiad—But Human Teenagers Still Come Out on Top

 

Date: July 29, 2025
Topic: AI at the International Mathematical Olympiad (IMO)

Johaan Joyson

In an unprecedented milestone, Google DeepMind’s Gemini Deep Think and OpenAI’s experimental model each solved five out of six IMO problems, scoring at the gold-medal level equivalent to 35/42 points under official grading standards. Google submitted Gemini to be independently graded by the IMO, while OpenAI's model was evaluated using the same criteria by external experts—both achieved the same score threshold. 

However, human contestants still outpaced the machines. Five high school students achieved perfect scores of 42 points, and several—including stars from the U.S. team—solved the most challenging combinatorics problem that even the AI famously missed.

Why It Matters:

1. AI Reasoning Enters New Territory

These AI systems showcased advanced reasoning: reasoning beyond pattern matching—multi-step logic, creative proof strategies, and symbolic deduction. This leap marks a turning point in the debate over whether AI can truly "think." 

2. Human Intuition Still Leads

While AI performed powerfully, humans still solved the toughest problems. Experts believe that deep combinatorial challenges and creative shortcuts remain uniquely human strengths—for now. 

3. Implications for Education

Educators are watching closely. AI systems like Gemini may soon serve as advanced tutors or reasoning assistants—but ensuring student critical thinking remains key. This could dramatically enhance fields from STEM to humanities.

4. Technology and Trust

To reach gold-level performance, both DeepMind and OpenAI used extended "thinking time", ensemble debate strategies, and careful selection of model outputs. These techniques highlight growing demands on compute and raise questions about transparency, fairness, and ethical training data. 

📝 Takeaway:

This milestone shows AI is now capable of top-tier logical reasoning on human problems—but it also underscores the continuing value of human creativity, insight, and intuition, especially in structured, abstract domains like mathematics. The future likely blends both: humans guiding insights and AI tackling complexity.