Google has started the rollout of Gemini 3 — the new generation of its AI models, which the company calls the most accurate and intelligent to date. For the first time, users get access to the flagship Gemini 3 Pro model directly in the Gemini app, as well as in “AI Mode” in Google Search for subscribers. The model processes text, images, and audio simultaneously, allowing for more complex tasks — such as turning recipe photos into a cookbook or creating interactive educational materials from video lectures.
Gemini 3 Pro provides improved coding, better understands queries, and forms concise, meaningful responses without empty compliments. The model is already available for users of the Gemini app in “Thinking” mode, and subscribers of Google AI Pro and Ultra in the USA can choose it for more complex tasks. For students in the USA, Google offers a year of free access to Google AI Pro, so they can try the new model without additional costs.
The Gemini app has received an updated interface with faster access to new chats and a “My Stuff” folder for created images, videos, and reports. Purchases are now integrated through Shopping Graph — prices, comparison tables, and offers from billions of products are displayed directly in the chat. Gemini 3 also opens a new class of generative interfaces — for example, the experimental “Visual Layout” mode turns a request for a three-day trip into a magazine with modules, and “Dynamic View” creates an interactive real-time guide.
For experienced users, the Gemini Agent feature is available, which connects to Gmail, Calendar, and other services to perform multi-step tasks. The agent can structure emails, create responses, or find car rentals using data from emails, and always asks for confirmation before performing important actions. Gemini Agent will initially launch for AI Ultra subscribers in the USA via the web interface.
Gemini 3 Pro leads performance rankings among modern AI models due to high scores on logic, math, and multimodal understanding tests. The Deep Think mode, which enhances the model’s ability for complex reasoning, is being tested by security experts and will be available later for AI Ultra subscribers. Google positions Gemini 3 as a foundation for learning, creation, and planning for both personal and corporate use.
