Google launches Gemini 3 with agents in its app

Google is updating the Gemini app with its most capable model yet, interfaces that change based on what you ask for, and an agent that can complete multistep tasks. Gemini 3 Pro is beginning its global rollout, while the most automated feature is initially reaching paid users in the United States.

Gemini 3 comes to the app

The main addition is Gemini 3, Google's new model for solving complex problems with more useful, better-organized and more concise responses. To try it, you need to select the “Thinking” option in the model selector.

Google also highlights its multimodal capabilities. That means it can interpret different types of information, such as text, images and audio, within the same conversation. For example, you can upload a photo of an exercise to ask for help or transcribe notes from a class you missed.

The model also improves Canvas, Gemini's workspace for creating applications and other projects through natural-language instructions. Google presents it as its best model yet for this type of assisted development, although the result will depend on what you ask for and how carefully you review what it generates.

Gemini 3 Pro has been distributed globally since November 18, 2025. Subscribers to Google AI Plus, Pro and Ultra keep higher usage limits. Google is also expanding one year of free access to Google AI Pro to college students in the United States.

Responses that become interfaces

The update introduces generative interfaces: responses that do more than display text and instead create a presentation adapted to your request. Google is starting with two experiments.

• Visual layout generates a magazine-style view with images and interactive modules.
• Dynamic view designs and programs a custom interface in real time to explain the topic you requested.

If you ask for a three-day trip to Rome, for example, Visual layout can show a visual itinerary that you can explore and adjust. If you ask about the works in the Van Gogh Gallery and their context, Dynamic view can build an interactive response with elements you can tap, scroll through and explore.

Both features are beginning to roll out, but at first you may see only one of them. Google presents them as experiments and is asking for feedback to compare how they work in practice.

The app is also introducing a simpler design, a folder called “My Stuff” for finding created images, videos and reports, and a redesigned shopping experience. The latter displays products, comparisons and prices using the Google Shopping Graph, which brings together more than 50 billion product listings.

Gemini Agent handles tasks for you

The most ambitious feature is Gemini Agent, an experimental agent that can break a complex request into multiple steps and carry them out within Gemini. It can connect to Google apps to manage your calendar, create reminders or organize your email.

An instruction such as “organize my inbox” could prompt it to prioritize tasks and prepare draft replies for you to review. You can also ask it to find a midsize SUV for a trip, use information from your email, compare options under 80 dollars per day and prepare the reservation.

The agent combines Gemini 3 with tools such as Deep Research, Canvas, web browsing and connected Google Workspace apps, including Gmail and Calendar. It should not send messages or make purchases without asking for your confirmation first, according to Google, and you can take control at any time.

For now, Gemini Agent is available on the web to Google AI Ultra subscribers in the United States. That limited scope matters: the feature is still experimental and does not amount to an autonomous assistant available to everyone.

For you, the main change is that Gemini is no longer just a conversation window. It can present information more visually, create tools adapted to each request and begin coordinating actions across your apps. The question to watch is whether these features maintain the accuracy and control needed when they move from demonstrations to real tasks involving your email, calendar and money.