Google is rolling out an updated version of Gemini 3 Deep Think mode in Gemini Apps for Google AI Ultra subscribers, focusing on further improving reasoning for complex tasks. According to the official introduction, this model explores a variety of hypothetical paths through the "advanced parallel thinking" mechanism, and then integrates and evaluates internally, so as to generate more robust problem-solving processes and answers in mathematical derivation, algorithm design and other scenarios.
This update continues the previous performance ofthe Gemini 2.5 Deep Think variant in the International Mathematical Olympiad (IMO) and the International University Programming Competition (ICPC) World Finals, extending the concept to the new generation of the Gemini 3 series. Compared with the conventional dialogue mode, Deep Think is more inclined to "slow thinking" and multiple rounds of internal deduction, which is suitable for dealing with complex reasoning, code generation, and scientific research problems, rather than simple Q&A or daily office tasks.
Currently, Ultra subscribers can enable this feature in the Gemini app by selecting "Deep Think" mode in the input box and checking Gemini 3 Pro in the model options. The official has not yet announced whether it will expand the scope of opening or adjust the call quota, but emphasized that Deep Think is still a high-order reasoning ability, and it needs to be exchanged for a certain amount of latency and computing power overhead for stronger thinking depth.
FAQs
Q: What is the difference between Gemini 3 Deep Think and regular Gemini 3 Pro?
A: Deep Think adds parallel hypothesis exploration and longer thinking paths to the reasoning process, prioritizing the accuracy of complex problems rather than response speed.
Q: Who can use Deep Think now?
A: Currently only available to users with a Google AI Ultra subscription, the corresponding options are visible in Gemini Apps.
Q: How do I enable Deep Think mode in the app?
A: Enable it by selecting "Deep Think" mode in the input box of the Gemini app and switching to Gemini 3 Pro in the model selector.
Q: What types of tasks is Deep Think suitable for?
A: It is especially suitable for complex math problems, algorithm and programming problems, scientific research reasoning, and scenarios that require multi-step logical analysis.
Q: How does this relate to the previous Gemini 2.5 Deep Think?
A: The new version is a capability upgrade based on the 2.5 Deep Think variant, continuing its deep reasoning ideas that have been verified in competitions such as IMO and ICPC.
