At this year’s International Math Olympiad (IMO), Google’s Gemini Deep Think AI competed under the same strict rules as human participants—and scored a gold medal, solving 5 out of 6 problems correctly. This marks a significant improvement over last year’s silver-medal performance by DeepMind’s AlphaProof + AlphaGeometry 2 system.
How Deep Think Outperformed Most Humans
- Natural Language Processing: Unlike previous AI models that required manual translation of problems into formal math notation, Deep Think understands and solves problems directly in natural language.
- Parallel Reasoning: Instead of following a single logical path, it explores multiple reasoning approaches simultaneously, improving accuracy.
- Rigorous Training: Google used reinforcement learning with long-form proofs—not just final answers—to ensure the AI shows its work, a key IMO requirement.
The One Problem It Missed
The sole incorrect answer came from Problem 3, which involved minimizing rectangle coverings. Deep Think started with a flawed assumption, leading it astray—yet only five human contestants solved it correctly.
Why the IMO Is a Tough AI Challenge
IMO problems test advanced algebra, combinatorics, geometry, and number theory—requiring creative, multi-disciplinary reasoning. Google’s success demonstrates:
- AI can now match top-tier human mathematicians in structured problem-solving.
- Natural language models can handle complex formal logic without manual intervention.
OpenAI’s Controversial Approach
While Google worked with IMO officials for official grading, OpenAI self-assessed its performance, claiming a gold medal without independent verification. Google’s Luong emphasized:
“We confirmed with the IMO that we solved five perfectly. Others might’ve lost a point and gotten silver.”
What’s Next for AI in Math?
- Gemini Deep Think will soon be available to Google AI Ultra subscribers ($250/month).
- DeepMind plans to refine the model for a perfect 6/6 score in future competitions.
The Verdict: AI Is Catching Up to Human Genius
Google’s achievement proves that AI can now compete at the highest levels of mathematical reasoning—though creativity and intuition still give humans an edge. With further advancements, IMO gold medals may soon be a regular feat for machines.
