Google has introduced Deep Think, a new feature for its Gemini app, available to Google AI Ultra subscribers starting August 1, 2025. This release is a variation of the Gemini 2.5 Deep Think model that achieved a gold-medal score at the 2025 International Mathematical Olympiad (IMO), solving five out of six problems for 35 out of 42 points. The consumer version is optimized for faster performance and daily usability, reaching a bronze-level performance on the 2025 IMO benchmark, according to Google's internal evaluations.

Key Features of Deep Think

1. Parallel Thinking: Explores multiple solution paths simultaneously to deliver thoughtful responses.
2. Extended Inference Time: Allows evaluation of hypotheses for complex problem-solving.
3. Reinforcement Learning: Improves problem-solving capabilities over time.
4. Tool Integration: Supports code execution and Google Search for enhanced functionality.

Performance Highlights

Outperforms models like Gemini 2.5 Pro and OpenAI’s o3 on:

1. LiveCodeBench V6 (coding performance).
2. Humanity’s Last Exam (science and math expertise).
3. Consumer version achieves bronze-level performance on the 2025 IMO benchmark.
4. Full IMO-grade model scored 35/42 points, solving 5/6 problems.

Use Cases

1. Web Development: Assists with iterative design and functionality improvements.
2. Scientific Research: Supports exploration of mathematical conjectures and complex literature.
3. Coding: Excels at problems requiring careful consideration of tradeoffs and time complexity.
4. Safety and Feedback
5. Improved content filtering and objectivity compared to Gemini 2.5 Pro.
6. May occasionally refuse benign requests.

Early feedback from testers, like mathematician Michel van Garrel, highlights its ability to generate proof ideas.

Availability

1. Google AI Ultra Subscribers: Access Deep Think in the Gemini app by selecting the 2.5 Pro model and toggling the Deep Think option.
2. Usage Limits: Restricted to a fixed number of prompts per day.
3. Research Access: Full IMO-grade model shared with select mathematicians and academics.
4. This release reflects Google’s ongoing efforts to enhance AI capabilities for complex problem-solving, supporting researchers, developers, and professionals in fields like mathematics and coding.