Google has announced the launch of its new Gemini 3 Deep Think mode within the Gemini application. This enhanced model functionality is exclusively available to Google AI Ultra subscribers and is designed for tasks requiring advanced logical processing.
According to Google, Deep Think ranks as the most effective model on the ARC-AGI-2 benchmark, which assesses complex reasoning abilities. The mode targets problem-solving in mathematics, natural sciences, and logic, which typically challenge even the most advanced models.
The rollout is currently underway, with the new feature gradually becoming available to users.