Google Gemini Rivals OpenAI by Achieving Gold Medal Performance at the International Mathematical Olympiad

AI Breakthrough at the International Mathematical Olympiad

The International Mathematical Olympiad (IMO) stands as the world’s premier stage for pre-university mathematicians, where elite student teams from around the globe confront six of the most challenging math problems imaginable. Traditionally, only about 8% of participants earn the coveted gold medal, marking them as top mathematical talents.

In a groundbreaking development, leading artificial intelligence innovators OpenAI and Google have pushed the boundaries of large language models (LLMs) by entering their latest AI models as virtual contenders in the 2025 IMO simulation. Both companies subjected their advanced LLMs to the Olympiad’s demanding standards: two separate 4.5-hour exam sessions, no access to external tools or the internet, and the requirement to produce natural-language proof solutions.

OpenAI Sets the Bar for AI Mathematical Reasoning

Earlier this week, OpenAI made headlines by announcing that its cutting-edge reasoning LLM matched the performance of top human contestants at the IMO. By solving Olympiad-level problems under authentic exam conditions, OpenAI’s model demonstrated an unprecedented ability to tackle complex theorem-proving and mathematical reasoning tasks—traditionally out of reach for even the most advanced AI systems. The company heralded this accomplishment as a major leap forward for artificial intelligence in mathematics.

 

Google Gemini Deep Think Surpasses Expectations

Following OpenAI’s bold claim, Google revealed that its enhanced Gemini Deep Think model also reached gold medal-level success by solving five out of six IMO problems. This latest iteration incorporates powerful advances in AI reasoning, leveraging Google DeepMind’s state-of-the-art research in parallel thinking. Unlike conventional models that process one solution at a time, Gemini explores multiple problem-solving strategies simultaneously, dramatically increasing efficiency and accuracy.

According to the Google DeepMind team, the new Gemini model was trained with novel reinforcement learning methods, allowing the AI to harness vast multi-step reasoning, sophisticated problem-solving, and advanced theorem-proving datasets. Additionally, Gemini was given access to an expertly curated repository of high-quality mathematical solutions, supplemented by strategic guidelines tailored to IMO-style problems.

Key Features and AI Innovations

• Parallel Thinking: Simultaneous exploration of multiple solutions for faster and more diverse problem-solving.
• Reinforcement Learning: Enhanced with cutting-edge techniques to master multi-step reasoning.
• Curated Training Data: Exposure to high-quality mathematical proofs and expert hints boosts the model’s competitiveness in mathematical domains.

AI vs AI: How Google Gemini Compares to OpenAI

Both Google Gemini Deep Think and OpenAI’s reasoning LLM have established themselves as world-class AI problem solvers, capable of matching, and in some areas, challenging top human competitors in mathematics. While OpenAI’s innovation showcases generalist reasoning at the Olympiad level, Gemini’s strength lies in its upgraded reasoning mode, parallel thinking, and reinforcement techniques, making it a formidable rival in academic AI benchmarks.

Potential Use Cases and Market Impact

The impressive performance of these LLMs at the IMO signals tremendous potential beyond academic competitions. Applications span from automated proof generation and advanced educational tools to assisting professional mathematicians with research. These breakthroughs could accelerate innovation in scientific research, STEM education, and beyond.

What’s Next for Google Gemini Deep Think?

Google has announced plans to release this advanced version of Gemini Deep Think to a select group of trusted testers in the near future. Following initial trials, the model will become available to Google AI Ultra subscribers, enabling lead users and developers to harness its remarkable problem-solving prowess.

This exciting race between OpenAI and Google signals a new era for artificial intelligence-driven discovery in mathematics, setting a gold-standard benchmark for future AI developments.