Google's I/O announcements: new models, a cloud agent that never sleeps, and a redesigned Gemini app

Google used its I/O developer conference to unveil a wave of new AI products. The highlights: a new model called Gemini 3.5 Flash, a multimodal model called Gemini Omni, and a personal agent named Gemini Spark that runs around the clock in the cloud. The Gemini app also gets a major refresh.

According to Google CEO Sundar Pichai, Gemini 3.5 Flash beats the roughly four-month-old Gemini 3.1 Pro on nearly every benchmark. The jump is especially sharp on GDP Val, a benchmark for economically relevant tasks, Google says. In an analysis by Artificial Analysis, Flash was the only model in the upper-right quadrant of intelligence versus speed—four times faster than other frontier models, according to Pichai.

An optimized version running on Google's in-house agent platform Antigravity is even twelve times faster, the company claims. Costs come in at roughly a third to half of comparable models, according to Google. Pichai did the math: companies that shift 80 percent of their workloads to a mix of 3.5 Flash and Pro could save over a billion dollars a year. Gemini 3.5 Pro is expected next month.

Chief AI Architect Koray Kavukcuoglu stressed that the 3.5 series was built for agentic work. The model can sustain autonomous sessions for several hours and run complex coding pipelines on its own, he said. Internally, Google had it build a working operating system from scratch as a test.

Google positions Gemini Omni as Veo's multimodal successor

Gemini Omni is Google's new model designed to produce any output from any input. Video ships first, with image and text coming later. Unlike Veo, which is a pure text-to-video model, Omni is built on the Gemini architecture and trained as multimodal from the ground up, according to Kavukcuoglu. The model can take its own generated output and feed it back in as input, allowing iterative editing, Google says.

Users can upload their own videos, swap out characters, or change the style. Asked whether Omni replaces Veo, Kavukcuoglu said Omni is a generalization of Veo. The path now leads consistently toward true multimodality, he added.

The first, faster variant named Omni Flash launches Tuesday for Google AI+, Pro, and Ultra subscribers in the Gemini app, in Flow, and in YouTube Shorts. An API version is coming, along with a more powerful Omni Pro, according to Google. All generated content carries Google's invisible SynthID watermark.

Google says Gemini Spark keeps working even when your laptop is closed

Gemini Spark is Google's personal agent for end users. According to Josh Woodward, who leads the Gemini app and AI Studio teams, it runs on dedicated virtual machines in Google Cloud, stays available 24/7, and keeps working in the background even when the user's device is off. Spark uses Gemini 3.5 with the full Antigravity pipeline for coding tasks, Google says.

Integration with Gmail, Docs, and other Workspace products is part of the launch. MCP connections to third-party services are coming in the next few weeks, according to the company. Chrome integration follows later, and on Android, a new UI surface called Android Halo will show what the agent is doing right at the top of the screen. Android Halo is expected later this year and will work not just with Spark but with other supported agents too, Google says. Devices running Gemini Nano will get extra features.

Spark rolls out this week to select testers and next week as a beta for Ultra subscribers in the US. Google is introducing a new Ultra plan at $100 per month and dropping the previous top-tier Ultra plan from $250 to $200.

Alongside Spark, a simpler agent called Daily Brief is coming to the app. It analyzes and prioritizes emails, calendar entries, and tasks overnight, according to Google. The feature builds on the Labs experiment "CC" from December.

Google says the Gemini app is getting its biggest visual overhaul yet

The Gemini app itself is getting what's probably the most visible update for end users. Google redesigned the app "from the ground up," according to Woodward. The new design language is called "Neural Expressive" and leans on fluid animations, new typography, and haptic feedback. Answers no longer show up as walls of text, Google says. Key information appears bold and at the top; scrolling reveals embedded images, timelines, or visualizations. Gemini Live now opens directly inline with no more switching between modes. The rollout starts today globally on Android, iOS, and the web, according to Google.

Gemini 3.5 Flash and Gemini Omni ship with the update. Through Omni, users can upload their own photos and videos from their gallery, apply pre-built templates, or drop themselves into generated scenes via an AI avatar, Google says. The app is now available in over 230 countries and more than 70 languages, according to Pichai. Monthly users grew from 400 million a year ago to over 900 million, and daily queries increased sevenfold, he said.

Google is also releasing a Gemini desktop app for macOS, available for download now. This summer, Spark will move into the desktop app, where it can access local files and automate workflows on the machine, according to the company. New voice features are planned for the summer too, turning free-form dictation directly into polished text drafts.

Google says Antigravity 2.0 gives developers a full agent orchestration platform

The developer platform Antigravity is getting a major update. Antigravity 2.0 is a standalone desktop app where users can orchestrate multiple autonomous agents in parallel, according to Google. It also comes with a CLI for terminal users and an SDK that provides the same agent harness Google uses to build its own products.

Internal growth has been massive, according to Pichai: in March, Google processed 500 billion tokens per day. That number now tops three trillion per day, he said. Google also introduced Codemender, a tool that uses Gemini reasoning to find vulnerabilities in code and patch them automatically, the company claims. Select experts get API access first, with a broader release to follow.

Google expands search, shopping, and watermarks

Search chief Liz Reid announced that AI Mode is switching to Gemini 3.5 Flash. The feature hit over a billion monthly users in its first year, according to Google; AI Overviews sits at 2.5 billion. The search box is being redesigned to accept longer, multimodal queries. In the coming months, users can build their own mini-apps within Search initially for AI Pro and Ultra subscribers in the US, Google says.

On the shopping side, Google is expanding the Universal Commerce Protocol (UCP) to cover hotels and delivery services. Amazon, Meta, Microsoft, Salesforce, and Stripe have joined the UCP Tech Council, according to the company. The Agent Payments Protocol (AP2) is designed to make sure agents only execute purchases within clearly defined limits.

On the transparency front, Google is extending SynthID to Search and Chrome. A right-click in the browser will let users check whether an image was AI-generated or edited, Google says. OpenAI, Kakao, and Levin Labs are joining the watermark standard as new partners, following NVIDIA, which signed on last year.

Asked whether competitors like Anthropic catching up to the frontier, like recently with Claude Mythos, had changed Google's self-image, Pichai gave a vague answer: the frontier isn't a fixed point where one provider stays ahead forever, he said. It shifts constantly. Sometimes one lab leads on certain benchmarks, sometimes another. Google leads across many dimensions and is focused on bringing frontier capabilities to as many users as possible, according to Pichai. A cheaper Flash model beating an older Pro model, he said, is exactly the point.