Google enables Gemini 3 Deep Think for subscribers
Google is adding Gemini 3 Deep Think to its app for AI Ultra subscribers. It can explore several hypotheses at once to tackle complex math, science, and logic problems.
Google now lets you use Gemini 3 Deep Think in its app, although for now it is available only to people with a Google AI Ultra subscription. The new mode is designed to solve complex math, science, and logic problems that require multiple steps of reasoning.
It is not a separate app or a model you need to install. In Gemini, choose Deep Think from the bar where you enter your request, then select Gemini 3 Pro from the model menu.
What makes Deep Think different
The mode uses parallel reasoning: instead of following a single path to reach an answer, it explores several hypotheses at the same time and compares the results. This can help when a question cannot be solved with a quick response, but requires checking calculations, ruling out alternatives, or following a long logical chain.
In practice, it is designed for tasks such as:
- Solving advanced math problems.
- Analyzing scientific questions with multiple conditions.
- Working through difficult riddles and logic problems.
Google says Gemini 3 Deep Think performs well on demanding tests such as Humanity’s Last Exam and ARC-AGI-2, two evaluations designed to measure reasoning abilities that go beyond answering typical questions.
The company also links this progress to variants of Gemini 2.5 Deep Think, which recently reached a level equivalent to a gold medal at the International Mathematical Olympiad and the world finals of the International Collegiate Programming Contest.
What changes for you
If you have Google AI Ultra, you can try it in the Gemini app starting today. The feature could be useful for academic problems, technical analysis, or tasks where a superficial answer is not enough.
For everyone else, the announcement does not provide immediate access. Google has not said in this update when Deep Think will reach other plans or whether it will offer a version with fewer capabilities.
The key question is whether this additional reasoning consistently improves answers in real-world situations, not just on exams designed to measure it. For now, Google is bringing that capability to its everyday product, but access remains limited to its highest-tier subscribers.