Google Unveils Gemini 3’s Deep Think Mode for AI Ultra Subscribers
Google has officially launched the Deep Think mode within its Gemini 3 platform, now accessible to Google AI Ultra subscribers. This advanced reasoning feature employs parallel reasoning techniques to explore multiple hypotheses simultaneously, leading to more refined and nuanced code outputs, particularly beneficial for complex visualizations and prototyping tasks.
Deep Think is designed to tackle intricate mathematical, scientific, and logical problems, with response times typically spanning a few minutes. Building upon the Gemini 2.5 version introduced in August, Deep Think has demonstrated significant improvements across various benchmarks:
– Humanity’s Last Exam: Achieved a 41.0% score without utilizing external tools.
– GPQA Diamond: Attained a 93.8% success rate.
– ARC-AGI-2: Recorded a 45.1% score with code execution, earning ARC Prize Verification.
In mid-November, Google announced a delay in the rollout of Deep Think to conduct thorough safety evaluations and gather feedback from safety testers. Following these assessments, the feature is now available to Google AI Ultra subscribers, who pay a monthly fee of $250. Subscribers can access Deep Think via the Tools menu by selecting the Thinking option in the model dropdown.
This release underscores Google’s commitment to advancing AI capabilities, providing users with sophisticated tools to address complex problem-solving scenarios.
