Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, information, and safety leaders. Subscribe Now
Google DeepMind introduced Monday that a complicated model of its Gemini synthetic intelligence mannequin has formally achieved gold medal-level efficiency on the Worldwide Mathematical Olympiad, fixing 5 of six exceptionally tough issues and incomes recognition as the primary AI system to obtain official gold-level grading from competitors organizers.
The victory advances the sphere of AI reasoning and places Google forward within the intensifying battle between tech giants constructing next-generation synthetic intelligence. Extra importantly, it demonstrates that AI can now deal with complicated mathematical issues utilizing pure language understanding slightly than requiring specialised programming languages.
“Official outcomes are in — Gemini achieved gold-medal stage within the Worldwide Mathematical Olympiad!” Demis Hassabis, CEO of Google DeepMind, wrote on social media platform X Monday morning. “A sophisticated model was in a position to remedy 5 out of 6 issues. Unbelievable progress.”
Official outcomes are in – Gemini achieved gold-medal stage within the Worldwide Mathematical Olympiad! ? A sophisticated model was in a position to remedy 5 out of 6 issues. Unbelievable progress – large congrats to @lmthang and the staff! https://t.co/pp9bXF7rVj
— Demis Hassabis (@demishassabis) July 21, 2025
The Worldwide Mathematical Olympiad, held yearly since 1959, is broadly thought of the world’s most prestigious arithmetic competitors for pre-university college students. Every taking part nation sends six elite younger mathematicians to compete in fixing six exceptionally difficult issues spanning algebra, combinatorics, geometry, and quantity concept. Solely about 8% of human contributors usually earn gold medals.
The AI Impression Sequence Returns to San Francisco – August 5
The following section of AI is right here – are you prepared? Be part of leaders from Block, GSK, and SAP for an unique have a look at how autonomous brokers are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.
Safe your spot now – house is proscribed: https://bit.ly/3GuuPLF
How Google DeepMind’s Gemini Deep Suppose cracked math’s hardest issues
Google’s newest success far exceeds its 2024 efficiency, when the corporate’s mixed AlphaProof and AlphaGeometry techniques earned silver medal standing by fixing 4 of six issues. That earlier system required human specialists to first translate pure language issues into domain-specific programming languages after which interpret the AI’s mathematical output.
This yr’s breakthrough got here by way of Gemini Deep Suppose, an enhanced reasoning system that employs what researchers name “parallel considering.” Not like conventional AI fashions that comply with a single chain of reasoning, Deep Suppose concurrently explores a number of potential options earlier than arriving at a last reply.
“Our mannequin operated end-to-end in pure language, producing rigorous mathematical proofs immediately from the official drawback descriptions,” Hassabis defined in a follow-up submit on the social media website X, emphasizing that the system accomplished its work throughout the competitors’s customary 4.5-hour time restrict.
We achieved this yr’s spectacular consequence utilizing a complicated model of Gemini Deep Suppose (an enhanced reasoning mode for complicated issues). Our mannequin operated end-to-end in pure language, producing rigorous mathematical proofs immediately from the official drawback descriptions –…
— Demis Hassabis (@demishassabis) July 21, 2025
The mannequin achieved 35 out of a potential 42 factors, comfortably exceeding the gold medal threshold. In response to IMO President Prof. Dr. Gregor Dolinar, the options had been “astonishing in lots of respects” and located to be “clear, exact and most of them simple to comply with” by competitors graders.
OpenAI faces backlash for bypassing official competitors guidelines
The announcement comes amid rising pressure within the AI trade over aggressive practices and transparency. Google DeepMind’s measured method to releasing its outcomes has drawn reward from the AI neighborhood, significantly in distinction to rival OpenAI’s dealing with of comparable achievements.
“We didn’t announce on Friday as a result of we revered the IMO Board’s unique request that every one AI labs share their outcomes solely after the official outcomes had been verified by unbiased specialists & the scholars had rightly acquired the acclamation they deserved,” Hassabis wrote, showing to reference OpenAI’s earlier announcement of its personal olympiad efficiency.
Btw as an apart, we didn’t announce on Friday as a result of we revered the IMO Board’s unique request that every one AI labs share their outcomes solely after the official outcomes had been verified by unbiased specialists & the scholars had rightly acquired the acclamation they deserved
— Demis Hassabis (@demishassabis) July 21, 2025
Social media customers had been fast to notice the excellence. “You see? OpenAI ignored the IMO request. Disgrace. No class. Straight up disrespect,” wrote one consumer. “Google DeepMind acted with integrity, aligned with humanity.”
The criticism stems from OpenAI’s determination to announce its personal mathematical olympiad outcomes with out taking part within the official IMO analysis course of. As an alternative, OpenAI had a panel of former IMO contributors grade its AI’s efficiency, a method that some in the neighborhood view as missing credibility.
“OpenAI is kind of probably the worst firm on the planet proper now,” wrote one critic, whereas others prompt the corporate must “take issues significantly” and “be extra credible.”
You see?
OpenAI ignored the IMO request. Disgrace. No class. Straight up disrespect.
Google DeepMind acted with integrity, aligned with humanity.
TRVTHNUKE pic.twitter.com/8LAOak6XUE
— NIK (@ns123abc) July 21, 2025
Contained in the coaching strategies that powered Gemini’s mathematical mastery
Google DeepMind’s success seems to stem from novel coaching strategies that transcend conventional approaches. The staff used superior reinforcement studying strategies designed to leverage multi-step reasoning, problem-solving, and theorem-proving information. The mannequin was additionally offered entry to a curated assortment of high-quality mathematical options and acquired particular steering on approaching IMO-style issues.
The technical achievement impressed AI researchers who famous its broader implications. “Not simply fixing math… however understanding language-described issues and making use of summary logic to novel instances,” wrote AI observer Elyss Wren. “This isn’t rote reminiscence — that is emergent cognition in movement.”
Ethan Mollick, a professor on the Wharton Faculty who research AI, emphasised the importance of utilizing a general-purpose mannequin slightly than specialised instruments. “Growing proof of the power of LLMs to generalize to novel drawback fixing,” he wrote, highlighting how this differs from earlier approaches that required specialised mathematical software program.
It wasn’t simply OpenAI.
Google additionally used a basic objective mannequin to unravel the very exhausting math issues of the Worldwide Math Olympiad in plain language. Final yr they used specialised device use
Growing proof of the power of LLMs to generalize to novel drawback fixing https://t.co/Ve72fFmx2b
— Ethan Mollick (@emollick) July 21, 2025
The mannequin demonstrated significantly spectacular reasoning in a single drawback the place many human opponents utilized graduate-level mathematical ideas. In response to DeepMind researcher Junehyuk Jung, Gemini “made a superb remark and used solely elementary quantity concept to create a self-contained proof,” discovering a extra elegant resolution than many human contributors.
What Google DeepMind’s victory means for the $200 billion AI race
The breakthrough comes at a essential second within the AI trade, the place firms are racing to reveal superior reasoning capabilities. The success has quick sensible implications: Google plans to make a model of this Deep Suppose mannequin accessible to mathematicians for testing earlier than rolling it out to Google AI Extremely subscribers, who pay $250 month-to-month for entry to the corporate’s most superior AI fashions.
The timing additionally highlights the intensifying competitors between main AI laboratories. Whereas Google celebrated its methodical, officially-verified method, the controversy surrounding OpenAI’s announcement displays broader tensions about transparency and credibility in AI growth.
This aggressive dynamic extends past simply mathematical reasoning. Latest weeks have seen numerous AI firms announce breakthrough capabilities, although not all have been acquired positively. Elon Musk’s xAI lately launched Grok 4, which the corporate claimed was the “smartest AI on the earth,” although leaderboard scores confirmed it trailing behind fashions from Google and OpenAI. Moreover, Grok has confronted criticism for controversial options together with sexualized AI companions and episodes of producing antisemitic content material.
The daybreak of AI that thinks like people—with real-world penalties
The mathematical olympiad victory goes past aggressive bragging rights. Gemini’s efficiency demonstrates that AI techniques can now match human-level reasoning in complicated duties requiring creativity, summary considering, and the power to synthesize insights throughout a number of domains.
“This can be a important advance over final yr’s breakthrough consequence,” the DeepMind staff famous of their technical announcement. The development from requiring specialised formal languages to working totally in pure language means that AI techniques have gotten extra intuitive and accessible.
For companies, this growth alerts that AI could quickly deal with complicated analytical issues throughout numerous industries with out requiring specialised programming or area experience. The power to cause by way of intricate challenges utilizing on a regular basis language may democratize subtle analytical capabilities throughout organizations.
Nevertheless, questions persist about whether or not these reasoning capabilities will translate successfully to messier real-world challenges. The mathematical olympiad supplies well-defined issues with clear success standards — a far cry from the ambiguous, multifaceted selections that outline most enterprise and scientific endeavors.
Google DeepMind plans to return to subsequent yr’s competitors “in quest of an ideal rating.” The corporate believes AI techniques combining pure language fluency with rigorous reasoning “will grow to be invaluable instruments for mathematicians, scientists, engineers, and researchers, serving to us advance human data on the trail to AGI.”
However maybe probably the most telling element emerged from the competitors itself: when confronted with the competition’s most tough drawback, Gemini began from an incorrect speculation and by no means recovered. Solely 5 human college students solved that drawback appropriately. In the long run, it appears, even gold medal-winning AI nonetheless has one thing to study from teenage mathematicians.
