Google has launched “Deep Think,” an advanced reasoning feature in the Gemini app for Ultra plan users. It lets the AI explore multiple ideas in parallel for better results on complex tasks like coding and math. Announced by Sundar Pichai on August 1, 2025, it’s now live in Gemini 2.5 Pro and available to developers via the Gemini API.
Let’s Know about Deep Think:-
- Deep Think is an advanced reasoning mode built into Gemini 2.5 Pro that enables the AI to consider multiple hypotheses in parallel before arriving at its final answer.
- It’s designed to enhance performance on complex, multi-step tasks like scientific reasoning, coding, and mathematical problem solving.
Availability & Access
- Deep Think is now live in the Gemini app for users of Google’s Ultra plan, which costs ~$250 /mo in the U.S.
- Subscribers can enable Deep Think via a toggle in the prompt bar when selecting the 2.5 Pro model.
- Wider access is planned via the Gemini API—initially to trusted testers and academics, which includes a special version that can reason for hours on complex problems.
Background & Context
- Deep Think first debuted at Google I/O 2025 in May, as part of enhancements to the Gemini 2.5 Pro series.
- Since then, Google has refined the model significantly before rolling it out to Ultra users today.
- Gemini app itself recently surpassed 450 million monthly active users, with India being a major source of growth thanks to local language support and app integration across Google services.
Summary Table
| Feature | Details |
|---|---|
| New tool | Deep Think, a multi-agent reasoning mode in Gemini 2.5 Pro |
| Launch date | August 1, 2025 (announced by Sundar Pichai) |
| Access | Google AI Ultra subscribers (~$250 /mo) via Gemini app |
| Key benchmark results | 87.6% on LiveCodeBench; gold-medal level performance on IMO benchmark |
| Developer access | Via Gemini API for trusted testers and academic partners |
