Google unveils Gemini 3.5 and agents at I/O 2026

Google is turning its search engine, apps and developer tools into a network of agents capable of acting for you. At I/O 2026, the company introduced Gemini 3.5 Flash, new video, shopping, email and programming features, along with plans for AI to work in the background with greater autonomy.

Most of the announcements are gradual rollouts or limited tests. Some are arriving this summer, while others remain available only to subscribers, testers or users in the United States.

Gemini 3.5 Flash aims to do more in less time

Google launched Gemini 3.5 Flash, a model that combines fast responses with the ability to solve complex tasks. The company says it outperforms Gemini 3.1 Pro in programming and agent tests, scoring 76.2% on Terminal-Bench 2.1, 1,656 Elo points in GDPval-AA and 83.6% on MCP Atlas.

The goal is to use it for work that previously took hours or days: building applications, maintaining code, preparing financial documents or coordinating tasks that require multiple steps. Google says it typically does this at less than half the cost of other advanced models, although that figure depends on the use case and the comparison selected.

The model is already available in Google Antigravity, the Gemini API, Google AI Studio and Android Studio. Google is also working on Gemini 3.5 Pro, which is expected to roll out next month.

Videos generated from any reference

Google introduced Gemini Omni, a family of models that starts by generating videos from text, images, videos or audio. Its goal is to maintain a coherent view of characters, scenes and movement, with a more precise understanding of elements such as gravity, kinetic energy and fluids.

In practice, you can upload a photo or video, describe the changes and request, for example, a cinematic zoom, a different background or a version in which you appear as an avatar. Gemini Omni Flash is beginning to roll out to the Gemini app and Google Flow for Google AI subscribers, as well as YouTube Shorts Remix and YouTube Create for users over 18.

Generated videos include an invisible digital watermark called SynthID. Google already lets you check for it in Gemini and is expanding verification to Search and Chrome.

Search goes beyond simply answering

Google says its AI search mode already has more than 1 billion monthly users and that queries in AI Mode have doubled every quarter since launch. Now, Gemini 3.5 Flash becomes its default model worldwide.

The new search box accepts text, images, files, videos and open tabs in Chrome. It also connects the general answers from AI Overviews with deeper conversations in AI Mode, without forcing you to start from scratch.

The company is also preparing information agents. You will be able to create several of them to monitor specific topics, such as changes in a flight's price, company news or sports results. These agents will check the web and real-time data, then send you summarized updates with possible actions. They will arrive first this summer for Google AI Pro and Ultra subscribers.

Another feature will let Search generate custom interfaces for a query: tables, charts, simulations or small dashboards. Later, you will be able to create mini-apps for ongoing tasks, such as organizing a move or planning a wedding.

AI that reviews your day and your shopping

Google introduced Gemini Spark, a personal agent that can work in the background on your phone or laptop. It is designed to act on your instructions and ask for confirmation before taking important actions. For now, it is available only to trusted testers, and the beta will first reach Ultra subscribers in the United States.

Daily Brief offers a more focused version of the same idea. During the night, it reviews your email, calendar and tasks, with your permission, to prepare a personalized summary of what you should address. It is rolling out to Google AI subscribers in the United States.

For shopping, Universal Cart brings together products you find in Search, Gemini, YouTube and Gmail. The cart can alert you to price drops, stock changes and incompatibilities, as well as compare offers, payment benefits and loyalty programs. Its rollout will begin this summer in Search and Gemini.

Antigravity turns programming into agent work

Google Antigravity 2.0 is a desktop application for coordinating several agents at once. One might program a website while another creates visual assets, for example. It also includes a terminal interface, a development kit and managed agents that can run code, handle files and browse the web inside isolated environments.

Google AI Studio lets you create Android apps with written instructions, test them in an emulator and publish them to Google Play's testing track. Beginners will be able to deploy their first two applications on Google Cloud at no cost and without a credit card.

The company also announced Google AI Ultra, a $100 per month plan with usage limits five times higher than AI Pro and 20 TB of storage. It is aimed at developers, technical teams and advanced creators.

Gmail, Docs and video get agents too

Gmail is adding AI Inbox, an inbox that prioritizes tasks and can draft replies based on the context of messages. Later, Gmail Live will let you ask questions about your email in a conversation.

Docs Live will turn a voice conversation into a structured document and, with authorization, can access Gmail, Drive, Chat and the web. In Keep, a similar feature will turn a "brain dump", or stream of ideas, into organized notes and lists.

In Google Flow, the new Flow Agent can plan audiovisual projects, create variations, edit multiple assets at once and organize files. You will also be able to describe custom tools in natural language to edit videos, create animations or modify images without writing code.

Science, learning and verifiable content

Gemini for Science brings together experimental tools for generating hypotheses, analyzing scientific literature and testing thousands of code variations. Google is gradually opening them starting May 19 through registration in Google Labs.

Ask YouTube will provide structured answers to complex questions and gather relevant long-form videos and Shorts. It will begin as an experiment this month for a group of desktop users searching in English from the United States.

Google is also preparing its first smart audio glasses with Gentle Monster, Warby Parker and Samsung. They will arrive in the fall and work with Android phones and iPhones.

The common thread at I/O 2026 is clear: Google wants AI not only to generate text or images, but also to monitor information, coordinate apps and complete tasks. The key question now is how much control users retain when these agents begin acting on their own, which features actually reach each country and what limits each subscription imposes.