Google has begun rolling out Gemini 3 Deep Think mode to Google AI Ultra subscribers. This is a new feature that the company claims delivers its most advanced reasoning capabilities to date. In a post on X (formerly Twitter), company CEO Sundar Pichai highlighted the model’s ability to handle highly complex queries. He tweeted that the new mode “brings our strongest reasoning capabilities yet,” and shared an example of its ability to “simulate complex 3D architecture.”
Gemini 3 Deep Think offers advanced reasoning for complex problems
According to Google, the new Deep Think mode is designed to tackle sophisticated challenges, including complex math, science and logic problems. The company adds that Deep Think operates by using iterative rounds of reasoning and advanced parallel reasoning to explore multiple hypotheses simultaneously.
“Deep Think is our most advanced reasoning mode that explores multiple hypotheses simultaneously to give you an even more sophisticated output. With Deep Think mode, Gemini uses iterative rounds of reasoning, resulting in more refined, nuanced code that helps you visualize, prototype and experiment with noticeably richer detail,” the company said.This method results in more refined and nuanced outputs, which the company says helps users “visualize, prototype and experiment with noticeably richer detail”, the company added.Google confirmed that Gemini 3 Deep Think is industry-leading on rigorous benchmarks, citing impressive results:Humanity’s Last Exam: Achieving 41.0% accuracy without the use of tools.ARC-AGI-2: Achieving an unprecedented 45.1% with code execution.These results build upon the success of previous Deep Think variants from Gemini 2.5, which recently reached a gold-medal standard at the International Mathematical Olympiad and the International Collegiate Programming Contest World Finals.Ultra subscribers can begin using Gemini 3 Deep Think mode immediately by selecting “Deep Think” in the prompt bar and Gemini 3 Pro in the model dropdown within the Gemini app.
