Google Introduces Deep Think Feature: Function Explained and Potential Reasons for Wider Accessibility
New Feature in Google's Gemini App Enhances AI Capabilities
Google has launched a new feature in its Gemini app, called Deep Think, aimed at improving the reasoning and problem-solving capabilities of the Gemini 2.5 Ultra model. This advanced multi-agent reasoning system simultaneously explores multiple ideas or hypotheses to produce the best answer.
Deep Think sets Gemini apart from its competitors by automatically integrating tools like code execution and Google Search, allowing it to perform complex tasks with greater accuracy and depth than previous models. The feature surpasses competitors such as OpenAI's GPT-4 and xAI's Gork 4 in key benchmarks, leveraging novel reinforcement learning and parallel thinking techniques to improve decision-making and problem-solving.
The Deep Think model works by spawning multiple AI agents that tackle questions in parallel, requiring more computational power but resulting in superior answers due to simultaneous consideration of various perspectives. The consumer-grade Gemini 2.5 Deep Think delivers responses in seconds to minutes with Bronze-level IMO performance, striking a balance between accuracy and speed for everyday use.
For Google AI Ultra subscribers, the Deep Think feature appears as an option in the Gemini app prompt bar when using the Gemini 2.5 Pro model, with limited daily prompts. Google plans to extend Deep Think access, including versions with and without additional tool integrations, to trusted developers via API soon.
If Deep Think proves successful and stable, it's likely that Google will broaden access to showcase its capabilities and stay competitive alongside other tech giants. It's worth noting that Deep Think is optimized for faster, more practical day-to-day use compared to the original research model.
Currently, Deep Think is only available to those who subscribe to the Gemini Ultra service, which costs $249 per month. It's also important to mention that Deep Think may sometimes refuse benign requests.
Google is actively performing frontier safety evaluations and working on mitigations as model capabilities advance. The company has already tested Deep Think with academic researchers, including mathematician Michel van Garrel, to explore its potential for advancing scientific discovery. Deep Think has shown strong performance on the 2025 International Mathematical Olympiad (IMO) benchmark, outperforming competitors on difficult reasoning tests.
In conclusion, Deep Think significantly enhances the reasoning and knowledge capabilities of the Gemini 2.5 Ultra model, making it the most powerful reasoning AI currently available from Google.
Google's Deep Think feature, present in the Gemini app, showcases an advancement in technology by leveraging AI to analyze problems and produce solutions more accurately and thoroughly. This technology sets the Gemini 2.5 Ultra model apart from competitors, such as OpenAI's GPT-4 and xAI's Gork 4, by integrating tools like code execution and Google Search, facilitating complex task performance.
