Google rolls out Gemini 3 Deep Think mode to Ultra users: Check what's new

Google has started rolling out its new Gemini 3 Deep Think mode to Google AI Ultra subscribers, offering major upgrades in reasoning, problem-solving, and complex task handling

Google has announced that it has begun rolling out the Gemini 3 Deep Think mode to Google AI Ultra subscribers within the Gemini app. According to the company, this new mode delivers enhanced reasoning capabilities and is specifically designed to handle complex maths, science and logic-based problems. Deep Think was previewed last month during the release of Google’s Gemini 3 AI models, but is only now becoming available to users.

Google Gemini 3 Deep Think: Details

Google says the new Gemini 3 Deep Think mode builds upon the earlier Gemini 2.5 Deep Think variants and uses “advanced parallel reasoning” to evaluate multiple hypotheses at once. This helps the AI assistant work through intricate equations, multi-step logic problems and other tasks that require deeper contextual understanding.

 

According to Google’s blog, Deep Think shows improvements in the following benchmarks:

• Humanity’s Last Exam: 41.0 per cent without the use of tools 
• ARC-AGI-2: 45.1 per cent with code execution

Google added that Ultra subscribers can try Gemini 3 Deep Think starting today by selecting “Deep Think” in the prompt bar and choosing Gemini 3 Pro in the model selector.

Google Gemini 3: Details

Google released its Gemini 3 AI model last month, bringing a major upgrade to the Gemini app, Search’s AI Mode and several Google services. As per Google’s blog, Gemini 3 has been engineered to provide more intelligent, more accurate and more natural responses than previous generations. The company said that the model delivers significantly deeper reasoning, enabling it to understand user intent more effectively, follow lengthy instructions with better reliability and break down complex tasks with fewer follow-up prompts.

Gemini 3 also features improved multimodal understanding, allowing it to process and interpret text, photos, screenshots, PDFs and videos with greater accuracy. This enables capabilities such as summarising long documents, analysing detailed images or explaining visual elements like graphs and charts.

 

Behind the scenes, Gemini 3 supports larger context windows and improved planning, helping it complete more extensive tasks — including organising information, generating structured plans or assisting with research — with reduced errors. Google also says Gemini 3 is its most secure model to date, with better defences against misleading prompts and more robust handling of sensitive or high-risk queries.