Google has released Deep Think, an enhanced AI upgrade for users of the Gemini app. This is an exclusive feature for Ultra subscribers only.
How is Deep Think going to be useful? Deep Think is based on a high-level model that previously achieved Bronze-level performance in the International Mathematical Olympiad (IMO), an Olympiad known for tough abstract math problems.
Previously, the IMO model took hours to solve difficult problems. Google has adjusted the model for fast, practical, every day use.
Although speed has improved, Deep Think retains its high-level reasoning ability concerning longer multi-step problem solving. It helps coders, researchers, and creators get the work done more efficiently.
Deep Think implements parallel thinking, checking multiple solution paths at the same time and is a significant part of how it enhances flexible thinking and outcome accuracy.
Parallel thinking enables the model to revise its approach, compare ideas and easily address more complex coding tasks and advanced research topics.
To take Deep Think for a test drive follow these instructions:
Open up the Gemini app on your mobile device
Select the 2.5 Pro model
Then toggle on Deep Think in the settings
Google has limited daily use at launch, but will soon allow developers and institutions to have wider access through the Gemini API.
Google’s CEO Sundar Pichai confirmed the release of Deep Think on X (formerly Twitter). He described it as a tool with “gold-medal performance at IMO.”
Pichai encouraged the user community to “take a look at Deep Think for complex coding, time complexity, deep scientific reasoning.” He said it’s for AI hobbyists.
According to Google, Deep Think has stronger safety filters and more accuracy than earlier generations. It even sometimes blocks entirely safe queries in the interest of safety.
Deep Think also offers more impartial responses and shows superior reliability in technical and academic use cases.
Deep Think is a major upgrade from the Gemini 2.5 demo highlighted at Google I/O 2025. It has now placed expert-level AI reasoning in the hands of users.
Google says they will be making their AI capabilities available to developers and academic institutions soon through the Gemini API.
This introductory phase of Deep Think confirms Google’s desire to make high-level AI reasoning available as a regular fixture of life for the average person around the world.