Google and OpenAI Model Wins Gold at International Math Olympiad

The International Math Olympiad is an elite mathematics competition for high school students which draws participants from over 100 countries. It was held in Australia this year, where a groundbreaking development was witnessed when Google DeepMind's Gemini Deep Think, an advanced version of their Gemini model, successfully solved five out of the six IMO problems perfectly, accumulating a total of 35 points. Parallel to this, OpenAI evaluated its experimental reasoning model on the same 2025 IMO problems. Later, researcher Alexander Wei shared OpenAI's results on social media on July 19 and revealed that their model also achieved 35 points, solving five out of six problems. 

1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO). pic.twitter.com/SG3k6EknaC

— Alexander Wei (@alexwei_) July 19, 2025

Both models adhered to the strict IMO rules, including two 4.5-hour exam sessions without internet access or other external tools, mirroring the conditions for human contestants. This achievement is particularly stunning given the difficulty of the IMO. This year, only 67 of the 630 human contestants (roughly 10%) received gold medals. Let's see how, through the scientific community, artificial intelligence models from (IMO) have emerged this past weekend. 

Check Out: What are the U.S. Time Zones? List of 6 U.S. Time Zones Explained

Dual Achievements in Advanced Math

OpenAI was not an official entrant in the competition, but its experimental reasoning model was evaluated on the same 2025 IMO problems. Researcher Alexander Wei shared OpenAI's results on social media on July 19, revealing that their model also achieved 35 points, solving five out of six problems. Both models adhered to the strict IMO rules, including two 4.5-hour exam sessions without internet access or other external tools, mirroring the conditions for human contestants.

This achievement is truly amazing because the IMO is very hard. This year, only about 10% of the human students won gold medals. That's just 67 out of 630 contestants. What makes the AI's success even more special is that these AI models are "general-purpose." This means they are designed to do many different things, like understanding language, writing computer code, and doing science. 

They are not like older AI systems, such as those that beat humans in games like Go or Poker. Those older systems were specially trained just for those specific games. The ability of these general-purpose models to tackle complex, abstract mathematical problems and deliver natural language proofs highlights a significant leap in AI's reasoning capabilities.

Implications for the Future of AI

Thang Luong is a senior research scientist at Google DeepMind, and he said that the AI model used in the competition is very similar to the main Gemini model. Moreover, he claimed that this is the model Google already offers to people. It means these advanced AI abilities might soon be available to more users. Mathematicians, especially, could use it to solve very difficult problems. OpenAI CEO Sam Altman echoed this sentiment, calling it "a significant marker of how far AI has come over the past decade."

Both companies were careful to respect the human element of the competition. For instance, Google waited for the IMO to officially certify its results before announcing them, out of respect for the student participants. Junehyuk Jung, who is an associate professor at Brown University and a visiting researcher at Google DeepMind, emphasised that the IMO's purpose is to promote the "beauty of mathematics" among high school students and encourage them to pursue the field.

Check Out: What is the new 'Visa Integrity Fee' for visiting the U.S.? Check Details Here

This unprecedented achievement signals a transformative moment for artificial intelligence, showing its evolving ability to master abstract thought and systematic reasoning. While not a "bots vs. humans" rivalry, this milestone undoubtedly opens new avenues for AI to assist and accelerate human discovery in mathematics and beyond, promising a future where AI acts as a powerful collaborator in pushing the boundaries of knowledge.