Google I/O 2026 developer conference: Extensive artificial intelligence upgrades

The most notable announcement was the Gemini 3.5 family of models, led by Gemini 3.5 Flash, a faster model designed for complex tasks, coding, multi-tasking and running agents. According to Google, the new model will be integrated by default in the Gemini app and in AI mode in the search engine. The company presents it as a model that is able to perform longer tasks, understand inputs of various types, such as text, images, video, audio and files, and act faster in the new generation of AI tools.

Watch: the full announcement video

Alongside Gemini 3.5 Flash, Google introduced Gemini Omni, a new model for video creation and editing. According to the company, Omni is designed to enable creation from videos, images, text and audio, and not just from a short verbal description. The user will be able to ask the system to create a new video, edit an existing video, change characters, change lighting, stabilize the camera or change elements in the scene by talking to the model. Google claims the model is also based on a better understanding of motion, gravity, kinetic energy and physical dynamics, to produce more realistic videos.

The new feature is expected to reach paid subscribers in the Gemini app and Google Flow first, initially in English. Meanwhile, Google also introduced Gemini Omni Flash, the first and faster version in the Omni family, which will be integrated into the company’s media creation tools. One of the more intriguing, and more sensitive, options is creating an AI avatar that looks and sounds like the user. Google emphasizes that every video created with the new models will be marked with SynthID, an invisible digital mark designed to help identify content created with artificial intelligence.

The second significant innovation is Gemini Spark, a personal AI agent that should perform tasks for the user. Unlike a regular chatbot, Spark is not content with answering a question, but should plan, organize, process information and operate within Google services. According to Google, he will be able, for example, to summarize emails, notes from meetings and chats, prepare a document with tasks to be performed, plan an event or vacation, and suggest follow-up actions. The company emphasizes that sensitive actions, such as sending an email or making a purchase, will require explicit authorization from the user.

Around the same idea, Google also announced a “daily summary” in the Gemini app. The feature is supposed to scan the user’s diary, tasks and emails every night, and build a snapshot of the coming day: what is urgent, what is important, what tasks require attention, and what can be done with the help of Gemini. In the first phase, the service will be available in the United States only, for users over the age of 18 and as part of Google AI subscriptions.

Google’s search engine is getting one of its biggest changes in years. The company introduced a new “AI search box”, designed to adapt to longer, more complex and conversational questions. Instead of briefly typing in keywords, users will be able to attach files, images, videos and even open tabs in the browser, and receive an answer based on the combination of the information. Google will try to refine the question, understand the intention and present a more adapted answer, alongside relevant links.

AI mode in search will move to run on Gemini 3.5 Flash, and Google presents it as faster and more suitable for complex searches. The company also announced Seamless Search, a smooth transition between AI mode and AI reviews, so that the search context is preserved and the results are updated accordingly. Simply put, Google wants the search to turn from a static text field into an ongoing conversation with an engine that knows how to remember what you’ve already searched for in that sequence and suggest the next step.

One of the areas where Google is trying to make search an actual action is shopping. Universal Cart, the new unified shopping cart, should bring together products from a variety of shopping sites and Google services. The user will be able to add an item he found in the search, in Gemini’s recommendation, and later also on YouTube or Gmail. The basket will track price drops, stock availability, benefits, price history and even product matching. In the example presented by Google, whoever assembles a computer will be able to receive an alert if one of the components he selected does not fit the rest of the system.

At the same time, Google announced Information Agents, agents that will monitor changing information in real time, such as flight prices, product availability, events, stocks or sports updates. Instead of returning again and again to the same search, the user will be able to define what interests him, and the agent will update him when there is a relevant change. In the next step, Agentic Bookings will allow agents to make reservations, such as concert tickets, restaurant tables and queues for various services, first in the United States and later in other areas.

The world of development also took center stage at the conference. Google Antigravity 2.0 was introduced as a desktop application for managing several AI agents at the same time, capable of performing complex tasks at the same time. The system has been upgraded with Gemini 3.5 Flash, and is mainly intended for developers, product teams and advanced users who want to run several work processes at the same time. Google AI Studio also received extensions, including a new mobile app, connection to the Google Workspace tool, Kotlin support for building Android apps using simple commands, and the ability to upload apps for testing and get initial feedback from users.

In the generative media segment, Google introduced Google Pics in Workspace, a tool for creating and editing images, flyers, presentations and infographics. Users will be able to change format, move objects, resize and edit images using text instructions. Google Flow, the company’s creative video tool, receives Flow Agent, a creative partner within the system, and Flow Tools, which allow you to build new creative tools, such as video effects or animations, using natural commands.

At DeepMind, Google presented GeoGenie, an extension to the Genie project, which allows you to select a real location in the world from Street View data and turn it into a 3D game environment in 360 degrees. The idea is based on technology that converts location to video and a virtual world, and opens the door for uses in games, simulations, virtual tourism and training.

The smart glasses also return to center stage. Android XR will be presented in two types of models: glasses with a voice interface directly to the ear, and glasses with a screen built into the lenses that displays information in real time. Gemini will be integrated into them as a personal assistant, which will be able to provide instructions, translation, information about the environment, messages and other actions without taking the phone out of the pocket. According to reports, the first models are expected to arrive in the fall, in collaboration with companies such as Samsung and eyeglass brands.

Google’s big challenge will not only be technological. As its agents gain access to the diary, email, shopping, personal preferences and information from linked applications, the need for trust on the part of users will increase. The question will no longer be only whether Gemini knows how to answer correctly, but whether users are willing to allow it to act on their behalf, track personal information, draft documents, offer shopping and manage part of their day. At the current conference, Google presented a very ambitious vision, but also one that will require it to prove transparency, security and caution as much as it proves speed and power.