TLDR: Google has officially launched its advanced AI model, Gemini 2.5 Deep Think, within the Gemini app. This model recently achieved a gold medal at the International Math Olympiad (IMO), marking a significant milestone as the first AI system to reach this level in the competition for high school students. While a version is available to Google AI Ultra subscribers, the full gold-winning model is being provided to a select group of mathematicians and academics for further research and development.
Google has announced the integration of its groundbreaking artificial intelligence model, Gemini 2.5 Deep Think, into the Gemini app. This strategic move, confirmed by Google CEO Sundar Pichai, follows the model’s remarkable achievement of a gold medal at the International Math Olympiad (IMO) competition, held from July 10-20 on Australia’s Sunshine Coast. This marks an unprecedented milestone, as it is the first time an AI system, including those from Google’s Gemini and OpenAI, has surpassed the gold-medal scoring threshold at the IMO for high school students .
The Gemini 2.5 Deep Think model, a variant of Google’s Gemini 2.5 large language model, is designed for complex problem-solving that demands strategic planning, creativity, and deep reasoning . A version of this powerful AI is now accessible to subscribers of Google’s AI Ultra plan, which costs $250 per month . This offering was initially teased at Google’s I/O developer conference earlier this year, with the promise that this model would justify the subscription cost .
For early feedback and further enhancement, the gold-winning Gemini 2.5 Deep Think model has been made available to a select group of mathematicians and academics . This model employs ‘parallel thinking’ techniques, enabling it to generate multiple ideas simultaneously and weigh potential solutions to intricate problems before arriving at a final answer. This approach aims to infuse creative thinking capabilities into the AI by exploring various hypotheses and providing it with ‘thinking time’ .
Despite its exceptional capabilities, the current version of the model can take hours to solve complex math problems . However, internal evaluations indicate that it is capable of achieving bronze-level performance on the 2025 IMO benchmark . The Gemini 2.5 Deep Think model has also demonstrated outstanding performance across various benchmarks, including LiveCodeBench V6, which assesses competitive code performance, and ‘humanity’s last exam,’ which tests expertise in diverse domains such as science and mathematics .
Also Read:
- Google Unveils Gemini 2.5 Deep Think: A Leap in Advanced AI Reasoning
- Google Revolutionizes Search with Advanced AI Features, Promising Enhanced User Experience and Business Growth in 2025
Looking ahead, Junehyuk Jung, a math professor at Brown University and visiting researcher in Google’s DeepMind unit, has expressed optimism about the future of AI in mathematics. He stated that AI is ‘less than a year away from being used by mathematicians to solve unsolved research problems’ . Pichai also noted the transformative potential of AI in this field . The latest release incorporates feedback from early trusted testers and research breakthroughs, representing a significant improvement over what was showcased at Google I/O earlier this year . While the earlier model took hours to reason about complex math problems, Google claims the latest model is faster and more usable for day-to-day activities .
