Google Deepmind announced on Monday that a sophisticated version of her Gemini Artificial Intelligence model was officially achieved Performance on the gold medal level on International Mathematical OlympicsSolution of 5 out of six exceptionally difficult problems and acquisition of recognition as the primary AI system that receives official gold level from the organizers of the competition.
The victory drives up the world of AI argument and puts Google on the intensifying struggle between tech giants, which construct up the synthetic intelligence of the following generation. It is much more necessary that AI can now tackle complex mathematical problems based on the understanding of natural language as a substitute of needing specialized programming languages.
“Official results have been in-Gemini on the international mathematical Olympics Gold Medal level!” Demis HassabisCEO from Google Deepmind, wrote on Monday morning on the Social Media platform X. “An advanced version was in a position to solve 5 out of 6 problems. Incredible progress.”
Official results are in -gini on the international mathematical Olympiad Gold Medal level! ? An advanced version was in a position to solve 5 out of 6 problems. Incredible progress – great congratulations @lmthang And the team! https://t.co/pp9bxf7rvj
– Demis Hassabus (@Demishabisabis) July 21, 2025
The International Mathematical OlympicsSince 1959, which takes place annually, the world's most respected mathematics competition for the world for college students in front of the university. Each participating country sends six young elite mathematicians to resolve six exceptionally difficult problems that include algebra, combination, geometry and number theory. Only about 8% of the human participants generally earn gold medals.
How Google Deepminds Gemini Deep think that probably the most difficult problems of mathematics have cracked
Google's latest success exceeds its performance of 2024 when the corporate is combined Alphapr And Alphageometry The systems received silver medal status by solving 4 out of six problems. This former system required human experts to first translate natural language problems into domain -specific programming languages after which interpret the mathematical edition of the AI.
This yr's breakthrough got here through Gemini Deep ThinkAn expanded argumentation system that the usage of what researchers call “parallel pondering. “” In contrast to traditional AI models that follow a single chain of argument, Deep Think at the identical time examines several possible solutions before they arrive to a final answer.
“Our model has operated end-to-end within the natural language and generated strict mathematical evidence directly from the official problem descriptions.” Hassabis explained In a follow-up contribution on the social media website X, the system emphasize that the system has accomplished its work throughout the 4.5-hour closing date of the competition.
We achieved the impressive results of this yr with a sophisticated version of Gemini Deep Think (an improved argumentation mode for complex problems). Our model operated end-to-end within the natural language and generated strict mathematical evidence directly from the official problem descriptions-…
– Demis Hassabus (@Demishabisabis) July 21, 2025
The model scored 35 of possible 42 points and conveniently exceeded the gold medal threshold. According to the IMO President Prof. Dr. Gregor Dolinar were the solutions “Amazing in some ways“And as” clear, precise and most of them easy to follow “of competitive levels.
Openaai is about setbacks to bypass the official competition rules
The announcement takes place under growing tension within the AI industry via competitive practices and transparency. The measured approach from Google Deepmind to publish its results praised the KI community, especially in contrast to competing Openai through similar achievements.
“We didn’t announce on Friday because we respected the unique request from the IMO board that each one KI laboratories only share their results after the official results were checked by independent experts and the scholars rightly received the attention they earned.” Hassabis wroteApparently to confer with Openais earlier announcement of their very own Olympiad performance.
BTW AS support, we didn’t announce on Friday
– Demis Hassabus (@Demishabisabis) July 21, 2025
Social media user quickly noticed the excellence. “You see? Openai ignored the IMO request. Shame. No class. Recommended, especially disrespectful”, ” wrote a user. “Google Deepmind acted with integrity that’s geared towards humanity.”
The criticism arises from the choice of Openaai to announce his own mathematical Olympic results without participating within the official IMO assessment process. Instead, Openaai had a bunch of former IMO participants who evaluated the performance of the AI, an approach that some locally thought to be an absence of credibility.
“Openai often is the worst company on the planet,” wrote a critic, while others suggested that the corporate “take things seriously” and “be more credible”.
Do you see?
Openai ignored the IMO request. Shame. No class. Respectiveness.
Google Deepmind acted with integrity and voted humanity.
Trvthnuke pic.twitter.com/8laoak6xue
– Nik (@ns123abc) July 21, 2025
Within the training methods that Gemini's mathematical championship have driven
Google Deepmind's success appears to be attributable to latest training techniques that transcend traditional approaches. The team used advanced reinforcement learning methods to make use of multi-stage argumentation, problem solving and theorem testing data. The model also received access to a curated collection of high-quality mathematical solutions and received specific guidelines for approaching the issues in IMO style.
The technical performance impressed AI researchers who found their broader effects. “Not only solve mathematics … but understand language -described problems and use abstract logic in latest cases,” wrote Ai Observer Elyss Wren. “This shouldn’t be a memory – that is an emerging perception in motion.”
Ethan MollickA professor on the Wharton School, who studies AI, emphasized the importance of using a general model and never using specialized tools. “Increased evidence of LLMS's ability to generalize itself to latest problems,” he wrote, underlining how this differs from previous approaches that required special mathematical software.
It wasn't just open.
Google also used an all -purpose model to resolve the very hard mathematical problems of the international mathematical Olympics in easy language. Last yr they used a special use of tools
Increasing evidence of LLMS's ability to generalize latest problem solving https://t.co/ve72ffmx2b
– Ethan Mollick (@emollick) July 21, 2025
The model showed a very impressive pondering in an issue through which many human competitors used mathematical concepts at graduate level. According to Deepmind Researcher Junehyuk Jung, Gemini has “made a superb statement and only used the elementary number theory to create a self-contained proof that found a more elegant solution than many human participants.
What the Google Deepmind victory for the AI race of 200 billion US dollars means
The breakthrough results in a critical moment within the AI industry, through which the corporate demonstrates superior arguments. Success has direct practical effects: Google plans to create a version of it Deep Think Model Available for mathematicians for testing before you run it to Google AI Ultra subscribers who pay 250 US dollars every month to access the corporate's most advanced AI models.
The timing also underlines the intensifying competition between large AI laboratories. While Google celebrated its methodical, officially verified approach, the controversy across the announcement of Openaai reflects broader tensions about transparency and credibility in AI development.
This competitive dynamic extends beyond the fair mathematical argument. In the past few weeks there have been various AI corporations that announce breakthrough functions, although not all of them were received positively. Elon Musks Xai was recently introduced Grok 4However, what the corporate claimed was the “most intelligent AI on this planet” The rating showed it after the conclusion Behind Google and Openai models. In addition, GROK has criticized to controversial characteristics including Sexualized AI companion and episodes of production anti -Semitic content.
The dawn of the AI, which like people thinks with real consequences
The Mathematical Olympics victory goes beyond competitive boast rights. Gemini's performance shows that AI systems can now correspond to the rationale on the human level in complex tasks that require creativity, abstract pondering and the flexibility to synthesis of data across several areas.
“This is a major progress in comparison with the breakthrough of last yr” Deepmind team noticed of their technical announcement. The progress from the demand from specialized formal languages to finish functionality within the natural language suggests that AI systems develop into more intuitive and accessible.
For corporations, this development signals that AI may soon have the opportunity to tackle complex analytical problems in various industries without special programming or domain expertise. The ability to argue through complicated challenges with on a regular basis language could democratize demanding analytical skills in organizations.
However, there are questions on whether these argumentation functions will effectively take care of the actual challenges of the actual world. The mathematical Olympics offers well-defined problems with clear success criteria away from the ambiguous, diverse decisions that outline most business and scientific efforts.
Google Deepmind plans to return to the competition of the following yr. “Looking for an ideal rating. “” “The company believes that AI systems that mix the flowing language with strict argument” for mathematicians, scientists, engineers and researchers develop into precious instruments and help us promote human knowledge “.
But possibly probably the most meaningful detail emerged from the competition itself: When Gemini was confronted with probably the most difficult problem within the competition, he began with a flawed hypothesis and never recovered. Only five human students properly solved this problem. In the top, even the AI with a gold medal still seems to learn something from teen mathematics.
