Google Rolls Out Gemini 3 Deep Think to Ultra Subscribers
Google started rolling out Gemini 3 Deep Think on December 4, 2025, to users on its top-tier AI plan. This new mode in the Gemini app focuses on tough reasoning tasks like math, science, and logic problems. It’s only for Google AI Ultra subscribers, who pay $250 a month, as noted in reports from 9to5Google and Mashable.
How to Get It
AI Ultra users can switch it on right away. Open the Gemini app, pick “Deep Think” in the prompt bar, and select Gemini 3 Pro from the model dropdown. Google’s official blog spells this out, and Tom’s Guide confirms it’s exclusive to that plan, which gives access to all Gemini features.
What Sets It Apart from Earlier Versions
Gemini 3 Deep Think builds directly on Gemini 2.5 Deep Think from August, which hit gold-medal levels at the International Mathematical Olympiad and International Collegiate Programming Contest World Finals. The big change is advanced parallel reasoning—it checks multiple hypotheses at once through iterative rounds. This leads to sharper outputs, like better code for visualizations or prototypes, according to 9to5Google.
Compared to standard Gemini 3 Pro, it pushes further on inference. GIGAZINE points out it beats Gemini 3 Pro and even GPT-5 Pro on tests measuring thinking and knowledge. Google delayed the launch briefly in November for safety checks, as mentioned in the 9to5Google piece.
Why It Stands Out: Benchmark Scores
It shines on hard tests that trip up other top models. From Google’s blog and echoed across sources:
- Humanity’s Last Exam: 41.0% without tools
- GPQA Diamond: 93.8%
- ARC-AGI-2: 45.1% with code execution (ARC Prize verified)
Mint calls these scores a new bar for AI reasoning, while Tom’s Guide explains how the parallel approach helps it crack logic puzzles faster than before.
Responses take a few minutes since it thinks deeper, making it ideal for pros handling tough work, not quick chats.
