Google's AI Ultra subscribers now have access to the new Deep Think mode, an advanced reasoning feature in Gemini 3. This mode allows the AI to explore multiple hypotheses simultaneously, resulting in more refined and nuanced code for better visualizations and prototyping. It's particularly useful for complex math, science, and logic problems, with answers typically provided within a few minutes.
Deep Think builds upon the Gemini 2.5 version, which was introduced in August. It has shown significant improvements in various benchmarks, including:
- Humanity's Last Exam: 41.0% without tools
- GPQA Diamond: 93.8%
- ARC-AGI-2: 45.1% with code execution, ARC Prize Verified
Google initially planned to launch Deep Think in mid-November but decided to delay it to ensure thorough safety evaluations and feedback from safety testers. This decision was likely influenced by the high demand for Gemini 3 Pro, as Google had to adjust the free access limits for the Pro version.
Google AI Ultra subscribers, who pay $250 per month, can access Deep Think from the Tools menu by selecting 'Thinking' in the model dropdown. Additionally, Google has launched Workspace Studio, allowing users to create automated Gemini-powered agents, and a web redesign for Gemini with a dark theme and a 'My Stuff' folder. The company also uses income-earning auto affiliate links, as disclosed in the FTC statement.
