Google launches Gemini 3 with Deep Think reasoning for complex tasks

Google DeepMind has unveiled Gemini 3, described as the company's most intelligent AI model to date. Building on advancements from Gemini 1 and 2, the new model combines native multimodality, long context windows, and agentic capabilities into a single system. It excels at grasping depth and nuance, understanding context and intent behind requests, and requires less prompting to deliver relevant results. Gemini 3 Pro sets new state-of-the-art performance in reasoning, coding, and multimodal benchmarks, while a forthcoming Deep Think mode will push intelligence further for complex problem-solving.

Starting today, Gemini 3 is shipping across Google's ecosystem: integrated into AI Mode in Search (the first time a new Gemini generation launches in Search on day one), the Gemini app, AI Studio, Vertex AI, and the new agentic development platform Google Antigravity. Sundar Pichai highlighted the scale of adoption, citing 2 billion monthly AI Overviews users, 650 million Gemini app users, and 13 million developers using Google's generative models. Gemini 3 is now available for developers and users, with Deep Think mode arriving soon for Ultra subscribers.