Google rolls out Deep Think in Gemini app: What it is, who can use, and more

Google Gemini Deep Think

Google is turning up the heat in the AI race with the launch of Deep Think — a powerful new reasoning feature now available to Gemini app users subscribed to the Google AI Ultra tier. Announced on August 1, Deep Think is designed to tackle complex problem-solving tasks using advanced parallel thinking techniques, enabling users to get more thoughtful, creative, and strategic responses from Gemini.

Deep Think builds on the version of Google’s Gemini 2.5 model that recently clinched a gold medal standard at the 2025 International Mathematical Olympiad (IMO). While the competition model required hours to solve intricate math problems, Deep Think is tuned for everyday usability, delivering Bronze-level IMO performance at faster response times, according to Google’s internal benchmarks.

At its core, Deep Think extends Gemini’s “thinking time,” allowing it to explore multiple hypotheses simultaneously and iteratively refine answers. Google says this parallel approach enables Gemini to tackle tasks like complex coding, scientific reasoning, algorithmic development, and creative design with greater depth and accuracy.

For developers and researchers, Deep Think could be a game-changer, says the company. Google has already shared the high-performance variant with a select group of mathematicians and academics to test its potential in formulating mathematical conjectures and navigating dense scientific literature.

The release also reflects major advances in reinforcement learning, enabling the AI to better navigate extended reasoning paths. This results in more intuitive problem-solving, especially in iterative workflows like web design, algorithm optimisation, and code development, where step-by-step improvements are critical.

Notably, Deep Think has demonstrated superior performance in challenging benchmarks like LiveCodeBench V6 and Humanity’s Last Exam, outpacing other models in coding and reasoning tasks without relying on external tools.

Google emphasised that safety remains central to Gemini’s development. Deep Think showed improved content safety and objectivity compared to Gemini 2.5 Pro, although it was more cautious, occasionally refusing benign prompts. Google is conducting extensive “frontier safety evaluations” to address risks associated with more advanced reasoning capabilities.

AI Ultra subscribers can access Deep Think in the Gemini app today, with a limited set of prompts per day. It integrates seamlessly with tools like code execution and Google Search, offering extended and detailed responses. Google also plans to release API access to trusted developers in the coming weeks.