Google introduced Gemini 3 with Deep Think and Antigravity support

Google officially introduced Gemini 3 — the new generation of its artificial intelligence model, distinguished by advanced reasoning capabilities, multimodality, and agentic behavior.

The model is capable of deeper analysis of queries, understanding user intent, and providing more structured responses. Gemini 3 processes text, images, video, and interactive simulations simultaneously, creating visual responses and tools tailored to user requests. In AI Mode, Google Search now uses "generative UI": visual layouts, sliders, tables, and simulations are dynamically generated based on the query. Alongside Gemini 3, Google is launching Antigravity — a development environment with multi-agent support, where AI can operate in the editor, terminal, and browser, generating plans, screenshots, logs, and fully documenting its work.

A new assistant function, Gemini Agent, has appeared, capable of performing multi-step tasks: for example, working with Gmail, the calendar, and planning projects. A separate deep reasoning mode, Deep Think, is also available, which is currently undergoing safety checks before a broad release. Gemini 3 Pro is available via API in Google AI Studio, integrates with Vertex AI, and the enterprise version, Gemini Enterprise.

The new model is already available in the Gemini app and will appear in Google Search in AI Mode for Pro and Ultra subscribers. Thanks to multimodality and visual interfaces, AI is becoming more interactive and useful in educational, research, and work scenarios, while tools like Antigravity open up new opportunities for developers and agentic applications, where AI not only responds to queries but actively acts and solves tasks.

Source: Google