Google Launches Gemini 3 with New Agent and Coding Interface

Google has released Gemini 3, its latest large language model, now available through the Gemini by Google app and AI search. The model delivers major improvements in reasoning and multimodal understanding, arriving seven months after Gemini 2.5. Google described it as its most capable model so far, with benchmark results setting new records for general reasoning and user satisfaction.

Gemini 3 introduces new features including generative interfaces that adapt to user prompts, producing dynamic visual layouts and interactive responses. A redesigned app interface makes it easier to organize content and access images, videos, and reports through a new “My Stuff” folder. The update also expands the model’s coding abilities, which Google refers to as “vibe coding,” enabling richer app development within the Gemini Canvas environment.

A new experimental tool, Gemini Agent, is being rolled out to Google AI Ultra members. It can perform multi-step tasks such as scheduling, inbox management, and research-based booking by connecting to Google apps. The agent is designed to execute complex instructions and streamline workflows directly within the Gemini interface.

Alongside the model release, Google introduced a Gemini-powered coding environment called Google Antigravity. The tool combines a chat-style prompt window, command-line interface, and integrated browser view to support agent-driven development. A research-oriented version of the model, Gemini 3 Deepthink, will follow after additional safety testing.