Humans succeed on AI in the annual mathematics Olympiad, but the machines are updating

Sydney -The humans beat the generative models made by Google and Openai in an international mathematics competition, but the programs reached gold scores for the first time, and the rate they are improving can cause some human introspection.

None of the AI models obtained complete grades, unlike five young people in the International Mathematical Olympiad (IMO), a prestigious annual competence where participants must be less than 20 years old.

Google said on Monday that an advanced version of his Chatbot Gemini had solved five of the six mathematical problems established in IMO, held in the Queensland of Australia this month.

“We can confirm that Google Deepmind has reached the very desired milestone, winning 35 of a possible 42 points, a gold medal score,” said the United States technological giant, cited the president of the IMO, Gregor Dolinar. “His solutions were amazing in many aspects. IMO students seemed clear, precise and most of them easy to follow.”

About 10% of human contestants won medals at gold level, and five received perfect 42 -point scores.

The US Chatgpt manufacturer. UU. Openai said that his experimental reasoning model had also obtained 35 gold level points in the test.

The result “achieved a great challenge in AI” in “the most prestigious mathematics competition in the world,” said Operai researcher Alexander Wei in a publication on social networks.

“We evaluate our models in the IMO problems of 2025 under the same rules as human contestants,” he said. “For each problem, three former IMO medalists independently described the presented test of the model.”

Google achieved a silver medal score in the IMO last year in the city of Bath, in the southwest of England, solving four of the six problems.

That took two or three days of calculation, much longer than this year, when its Gemini model solved the problems within the 4.5 -hour time limit, he said.

The IMO said that technological companies had “proved privately the code models closed in this year’s problems”, which face 641 students in the competence of 112 countries.

“It is very exciting to see the progress in the mathematical capabilities of AI models,” said the president of Imo, Dolinar.

The organizers of the contest could not verify how much computer power had used the AI models or if there had been human participation, he said.

In a Interview with the 60 minutes of News Earlier this year, one of Google’s main researchers predicted that in just five to 10 years, computers that have cognitive skills at a human level would be made, a milestone known as “artificial general intelligence.”

Google Deepmind CEO Demis Hassabis predicted That AI technology was on the way to understanding the world in a nuanced way and not only to solve important problems, but even to develop a sense of imagination, within a decade, thanks to an increase in investment.

“It is moving incredibly fast,” said Hassabis. “I think we are in some kind of exponential improvement curve. Of course, the success of the field in recent years has attracted even more attention, more resources, more talent. So that adds to this exponential progress.”

• Google
• Technology
• Opadai
• Robot
• Artificial intelligence
• Science
• Chatgpt