Google unveils Gemini AI push across Search & apps

Google unveiled a series of new Gemini artificial intelligence products and model updates at its I/O developer conference, extending its push to embed AI agents and conversational tools across Search, YouTube, Workspace and its developer offerings.

The updates span consumer software, cloud infrastructure and AI model development. Google presented them as the next stage of its Gemini strategy, with Chief Executive Officer Sundar Pichai emphasising practical uses for AI in everyday products.

Among the most prominent announcements was Gemini 3.5 Flash, a new model that combines improved reasoning and coding with faster responses. It is available across Google's products and application programming interfaces, and Google described it as the first in a series built to take action as well as generate responses.

Google also introduced Gemini Omni, a new model family designed to generate outputs across different media types from a range of inputs. The first version, Gemini Omni Flash, starts with video output and is being made available in the Gemini app, Google Flow and YouTube Shorts, with access for developers and business customers to follow.

Search changes

Search featured heavily in the rollout. AI Overviews now has more than 2.5 billion monthly active users, while AI Mode has passed 1 billion monthly active users within a year.

Google is adding what it calls information agents to Search, allowing users to set background AI tools to monitor specific information and act when needed. It also plans to add generative user interface features that build custom layouts and visuals in response to individual queries, along with persistent dashboards and trackers for longer-running tasks.

The additions reflect a broader effort to make Search less a list of links and more an ongoing interaction. Some of the new Search functions will begin with subscribers to Google's AI Pro and AI Ultra services, while other generative interface tools will be offered more widely.

Consumer tools

In the Gemini app, Google announced Gemini Spark, a personal AI agent designed to carry out longer tasks in the background on behalf of users. The tool runs on dedicated virtual machines in Google Cloud and is intended to work through the Gemini app, with email, chat, Android and Chrome integration also planned.

Spark is being rolled out first to a limited group of testers, followed by a wider beta for certain paying users in the United States. Another new feature, Daily Brief, will provide a personalised digest based on data from inboxes, calendars and task lists.

YouTube is also getting a feature called Ask YouTube, intended to answer user questions by identifying relevant videos and jumping to the most relevant point in a clip. In Workspace, Google announced Docs Live, a voice-driven feature that lets users speak their ideas aloud and have Gemini turn them into a document.

The Gemini app has grown from 400 million monthly active users last year to more than 900 million. Daily requests in the app have increased more than sevenfold over the same period.

Developer push

For developers, Google is expanding Antigravity, its internal and external platform for AI-driven software creation. Antigravity 2.0 will include a standalone desktop application designed to manage multiple autonomous agents.

More than 8.5 million developers are now building applications and services with Google's models each month. Its model APIs are processing about 19 billion tokens per minute, while more than 375 Google Cloud customers each processed more than one trillion tokens over the past year.

Pichai highlighted the growth in computing demand across Google's products, saying monthly tokens processed across the company's services have risen to more than 3.2 quadrillion. He also said Google expects annual capital expenditure to reach about USD $180 billion to USD $190 billion this year, up from USD $31 billion in 2022.

Infrastructure and standards

Part of that spending is going toward Google's in-house tensor processing units. Its eighth-generation TPU line now uses separate designs for training and inference, branded TPU 8t and TPU 8i. Google said the new chips are intended to improve speed and energy efficiency as AI workloads increase.

Google also sought to address concerns over AI-generated media. Its SynthID watermarking system has now marked more than 100 billion images and videos, alongside 60,000 years of audio assets. The company is extending content verification tools across Search and Chrome.

OpenAI, Kakao and Eleven Labs are adopting SynthID, joining Nvidia. The move signals a degree of cross-industry cooperation on identifying AI-generated content as synthetic media becomes harder for users to detect.

Other announcements included Google Pics, an AI image creation and editing tool, updates to Google Flow for creative work, and further details on AI-enabled eyewear. Google also outlined Gemini for Science, a package of AI tools aimed at research workflows and connections to more than 30 life sciences databases and tools.

"Over 8.5 million developers are now building new apps and experiences with our models monthly," Pichai said.