New Delhi: Google has introduced Gemini 3 Deep Think, a new reasoning mode designed to strengthen the model’s ability to handle complex, multi-step analytical tasks. The feature is available to Google AI Ultra subscribers through the Gemini app.
Deep Think applies a parallel reasoning approach that evaluates multiple hypotheses before selecting an outcome, enabling the model to maintain context across longer problem sequences. Google positions this update as part of its effort to shift AI systems from short, conversational outputs toward tools that can support research-oriented and technical workloads.
The company reports notable gains on established benchmarks. Gemini 3 Deep Think scored 41% on the “Humanity’s Last Exam” advanced reasoning test and 45.1% on ARC-AGI-2 when paired with code execution, both considered strong indicators of machine reasoning improvements. These results extend progress earlier demonstrated by Gemini 2.5 Deep Think.
Users can access the new mode by activating the “Deep Think” option while using Gemini 3 Pro in the app. It supports everyday queries and more specialised analytical tasks across devices.
Analysts say the launch reflects intensifying competition among major AI developers to enhance reasoning capabilities, a domain increasingly seen as critical for scientific, engineering and professional applications.
Also Read –
