- By Alex David
- Mon, 08 Dec 2025 12:48 PM (IST)
- Source:JND
Google has officially rolled out its next-generation reasoning model, Gemini 3 Deep Think, to eligible users. First teased during the global release of Gemini 3 Pro last month, the new model arrives just weeks later and is positioned as Google’s most capable artificial intelligence system to date. Deep Think is designed to handle extremely complex and niche problems that demand advanced logic, mathematical reasoning, and scientific understanding. However, access is intentionally limited — the model is available only to users subscribed to Google’s most expensive AI Ultra plan.
Gemini 3 Deep Think Availability and Access
In a blog post announcing the launch, Google confirmed that Gemini 3 Deep Think is restricted to Google AI Ultra subscribers, a plan priced at around Rs. 24,500 per month in India. Users can access the model via the Gemini app or website by selecting Gemini 3 Pro in the model picker and enabling the Deep Think option directly from the prompt bar.
ALSO READ: Realme Narzo 90 Series 5G Launch Date Teased In India: Check Out Everything We Know So Far
Google says this controlled rollout is intentional, allowing the company to monitor performance closely while offering its most advanced reasoning tools to power users, researchers, and professionals working on high-complexity tasks.
Why Gemini 3 Deep Think Is a Big Deal
Interest around Deep Think has been high ever since Google revealed early benchmark claims last month. The model has reportedly posted record-breaking scores on some of the toughest reasoning tests used in AI research.
Recommended For You
On Humanity’s Last Exam, a benchmark focused on deep academic reasoning, Gemini 3 Deep Think scored 41 percent without using any external tools. Google claims this is the highest score ever achieved on the test, putting the model well ahead of current competitors.
The model also performed strongly on ARC-AGI-2, a notoriously difficult benchmark involving abstract visual puzzles that require code execution to solve. Gemini 3 Deep Think achieved a 45.1 percent score, which Google described as an unprecedented result. The ARC-AGI-2 test is often used as an indicator of how close an AI system may be to artificial general intelligence.
How Gemini 3 Deep Think Works
The improvement in the performance of the model is primarily due to a new way of thinking known as "parallel reasoning". In contrast to examining a single path of thinking that leads toward a solution to a problem, the Deep Think will examine numerous possible paths simultaneously, compare each solution path with the others and hone in on the best answer through continual internal reviews. By using these internal evaluations, Deep Think can explore and assess ideas and make high-quality decisions regarding how to address very complicated questions.
Final Thoughts
With Gemini 3 Deep Think, Google is clearly pushing the boundaries of AI reasoning. While the high subscription cost limits accessibility, the model signals where top-tier AI systems are headed — toward deeper reasoning, higher reliability, and more advanced problem-solving. For now, Gemini 3 Deep Think stands as Google’s strongest response yet to rising competition from OpenAI and Anthropic.