Google enables 'Deep Think' mode in Gemini: What is it, how it works, more

Google said Gemini 2.5's Deep Think mode brings human-like reasoning to AI, enabling it to generate multiple ideas in parallel, explore hypotheses, and arrive at more creative solutions

Google is rolling out the Gemini 2.5 Deep Think mode to AI Ultra subscribers. Previewed at the I/O 2025 conference, Deep Think in Gemini can generate multiple, “parallel streams of thought” simultaneously. According to Google, this parallel thinking technique enables Gemini to generate many ideas at once and consider them side by side — similar to how people think — before refining, combining, or revising them to arrive at the best possible answer.

 

Additionally, Deep Think mode extends “thinking time,” allowing the AI to explore various hypotheses before settling on “creative solutions” to complex problems.

 

What Gemini Deep Think mode can do

 

 

Google says Deep Think is designed for tasks that demand creativity, strategic thinking, and a step-by-step refinement process. In one demonstration, the company asked Gemini to generate HTML code for a detailed animation of a hyper-realistic pagoda. When comparing the outputs from different Gemini models, the Deep Think version delivered a more visually appealing result with greater design complexity and finer details.

 

According to Google, Deep Think can assist users in several key areas:

• Enhancing both visual appeal and functionality in web development.
• Generating and evaluating mathematical conjectures or interpreting dense scientific research.
• Solving complex programming challenges that require strategic planning and an understanding of trade-offs like time complexity.

 

Availability of Gemini 2.5 Deep Think mode

 

Gemini 2.5 Deep Think mode is now live on the Gemini mobile app and web interface for Google AI Ultra subscribers. To activate it, users must choose the 2.5 Pro model in the drop-down menu and toggle Deep Think from the prompt bar. Once enabled, it integrates automatically with features like code execution and Google Search.

 

The company also said that it is working on releasing Deep Think with and without tools to a set of trusted testers via the Gemini API in the coming weeks.