This year at Google I/O 2025, it was announced that Gemini 2.5 Pro would be added.Deep Think function, now officially open for subscriptionGoogle AI Ultra SolutionThe model is used by users and will think deeply before answering the user's questions.
Results approach the gold medal standard of the International Mathematical Olympiad, strengthening "parallel thinking" reasoning skills
According to Google, Deep Think's core advantage lies in its ability to engage in "parallel thinking" simultaneously. This allows the AI to develop multiple thinking paths simultaneously when responding to user commands, and to provide the best answer after analysis and cross-evaluation, significantly improving the accuracy and efficiency of AI in logical reasoning, multi-step inference, and data integration.
In previous internal testing, Deep Think even achieved gold medal performance at the International Mathematical Olympiad. The currently available version has undergone minor adjustments to enhance practicality and stability. While it only achieves bronze medal performance, its response speed and accuracy in everyday use are already quite impressive. Google stated that the original gold medal version will remain available to select researchers and mathematician teams for research purposes.
Enhanced multimodal capabilities to support understanding of text, images, and audio data
In addition to enhancing its logical reasoning capabilities, Deep Think also fully supports Google's multimodal input feature, which has been a major focus in recent years. Users can input data in various forms, such as text, images, and audio, allowing the AI to integrate and analyze data and provide corresponding suggestions or conclusions. This makes Deep Think suitable not only for use in technology and scientific research fields, but also for diverse applications such as video creation and educational assistance.
Google stated that Deep Think performed well in multiple AI capability assessments, including the "Humanity's Last Exam," which tests users with diverse problem scenarios, and the real-time code application test "LiveCodeBench." Its results surpassed the existing Gemini 2.5 Pro version, and also beat OpenAI's o3 model and Grok 4, launched by xAI, founded by Elon Musk.
Currently only available to Google AI Ultra users, with a daily interaction limit
Deep Think is currently only available to users with Google AI Ultra subscriptions and has been integrated into the Gemini app, allowing users to experience it on mobile devices or the web. However, it's important to note that even with the higher-tier subscription, Google still caps the number of interactions with Deep Think per day (the specific number hasn't been disclosed), highlighting the model's resource-intensive nature.
Google AI Ultra subscriptions cost $249.99 per month and, in addition to access to Deep Think, also include additional computing resources, priority testing permissions, and full access to multimodal models. This upgrade provides more flexible and powerful tools for professional users with AI research needs, development tasks, or those looking to further enhance their decision-making and analytical capabilities.
AI competition continues to heat up, with Google, OpenAI, and xAI competing against each other in advanced model development.
As AI model applications gradually move from text generation to reasoning and integrated analysis, technology giants including Google, OpenAI, and xAI are investing resources in the development of multimodal and strong logic models.
Although the price and interaction limitations may deter ordinary users, judging by Google's emphasis on the accuracy and practicality of Deep Think, this advanced model has become an important milestone in the entry of AI applications into the "professional assistant" stage.
