Google has launched Gemini 2.5 Deep Think, a new AI model designed for complex problem-solving, which is now available to subscribers of its Google AI Ultra plan.
The company states that the model uses a technique called “parallel thinking.” This approach allows the AI to generate and consider many different ideas or solution paths at the same time before delivering a final answer. Google says this results in more detailed and creative responses, particularly for tasks that require step-by-step planning and improvement.
It is important to note that the publicly available model is a faster, more accessible variation of the AI that achieved a gold-medal standard at the International Mathematical Olympiad (IMO). According to Google, this subscriber version reaches “Bronze-level performance” on the same benchmark. The full, slower “gold” model has been provided to a select group of mathematicians for research purposes.
Key Applications and Performance
Google highlights the model’s effectiveness in several areas:
- Scientific and mathematical discovery: Helping researchers formulate and explore complex conjectures.
- Algorithmic development: Excelling at difficult coding problems where careful planning is crucial.
- Iterative design: Improving the functionality and aesthetics of projects like web development.
The company also reports that Gemini 2.5 Deep Think achieves state-of-the-art results on challenging industry benchmarks that measure coding and reasoning, outperforming several competing models. Access is provided through a toggle in the Gemini app for Ultra subscribers, who have a fixed number of “Deep Think” prompts per day.
Sources: Google, TechCrunch, VentureBeat