Google I/O 2026: Gemini 3.5 Flash, Spark Agent, and Omni Model

Key Takeaways

• Gemini 3.5 Flash is four times faster than other frontier models and costs a third to half as much
• Gemini Spark is a personal AI agent that runs continuously in the cloud for hours-long autonomous sessions
• Gemini Omni handles video, image, and text generation from any input type

Google's I/O 2026 developer conference delivered a stack of AI announcements. The company introduced three new Gemini products: a faster and cheaper model called Gemini 3.5 Flash, a multimodal system named Gemini Omni, and a personal AI agent called Gemini Spark that runs around the clock in the cloud.

The Gemini app also received a visual overhaul, and Google updated its Antigravity platform for managing autonomous agents. Here's what each announcement means in practice.

Gemini 3.5 Flash: Speed and Cost Are the Pitch

CEO Sundar Pichai said Gemini 3.5 Flash beats the four-month-old Gemini 3.1 Pro on nearly every benchmark. The sharpest improvement appeared on GDP Val, a benchmark measuring economically relevant tasks.

According to an Artificial Analysis study cited by Google, Flash was the only model in the upper-right quadrant of intelligence versus speed. Pichai claimed it runs four times faster than other frontier models. An optimized version on Google's in-house Antigravity platform hits twelve times faster, the company says.

Pricing sits at roughly a third to half of comparable models. Pichai did the math publicly: 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.

Built for Agentic Work

Google Chief AI Architect Koray Kavukcuoglu said the 3.5 series was designed for agentic work from the start. The model can sustain autonomous sessions for several hours and run complex coding pipelines without human intervention.

As a stress test, Google had the model build a working operating system from scratch internally. That's a notable claim, though the company didn't share details on the OS's complexity or functionality.

Gemini Omni: Any Input, Any Output

Gemini Omni is Google's new multimodal model designed to produce any output from any input. Video generation ships first, with image and text capabilities coming later.

Unlike Veo, Google's 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 as input, allowing iterative editing. Users can upload videos, swap characters, or change visual styles.

When asked whether Omni replaces Veo, Kavukcuoglu said Omni is a generalization of Veo. The path now leads toward true multimodality, he added.

The first variant, Omni Flash, launches Tuesday for Google AI+, Pro, and Ultra subscribers. It will be available in the Gemini app, in Flow, and in YouTube Shorts. An API version is coming.

Gemini Spark: A Cloud Agent That Never Logs Off

Gemini Spark is Google's personal AI agent. Unlike chatbots that respond to prompts and then idle, Spark runs continuously in the cloud. It's designed for tasks that take hours, not seconds.

Google didn't detail specific use cases in the announcement, but the combination of the 3.5 series' agentic capabilities with persistent cloud execution suggests applications like long-running research, code generation pipelines, and automated workflows that don't require constant human check-ins.

Antigravity Gets Multi-Agent Management

For developers, Google updated its Antigravity platform to manage multiple autonomous agents. The platform handles orchestration, and the optimized Flash model running on Antigravity hits the twelve-times-faster benchmark Google cited.

Google also expanded its AI search features and the SynthID watermarking tool, though details on these updates were lighter in the announcement.

The Gemini App Redesign

The Gemini app received a major visual refresh. Google didn't elaborate on specific interface changes, but the redesign arrives alongside the new model capabilities and Omni Flash access for subscribers.

Frequently Asked Questions

What is Gemini 3.5 Flash?

Gemini 3.5 Flash is Google's new AI model designed for speed and cost efficiency. It runs four times faster than comparable frontier models and costs a third to half as much, according to Google.

What is Gemini Spark?

Gemini Spark is a personal AI agent that runs continuously in the cloud. It's built for tasks that take hours, maintaining autonomous sessions without requiring constant human prompts.

How is Gemini Omni different from Veo?

Veo is a text-to-video model. Gemini Omni is multimodal from the ground up, handling video, image, and text generation from any input type. Google describes Omni as a generalization of Veo.

When does Gemini 3.5 Pro release?

Google said Gemini 3.5 Pro is expected next month, following the Flash release.

What is Antigravity?

Antigravity is Google's in-house platform for managing autonomous AI agents. The updated version supports multiple agents and runs an optimized Flash model at twelve times the standard speed.

Source: The Decoder / Maximilian Schreiner

Gemini Omni: A New 'World Model' for Physical Simulation and Video Editing

The new article reveals that Gemini Omni functions as a 'world model' designed to simulate physical concepts like gravity and kinetic energy. It also details specific capabilities like natural language video editing and the model's architecture, which integrates reasoning from Google's Veo and Genie media models.

Google Search Integration and AI Mode Enhancements

The new article provides specific details on Gemini 3.5 Flash's integration into Google Search's 'AI Mode,' including a dynamic expanding text box and support for Chrome tabs as a search modality. It also introduces 'Personal Intelligence' features for localized search results across 200 countries and the ability to manage multiple AI agents directly within the search interface.

AI-Powered Google Search Overhaul

The new article details a significant overhaul of Google Search, including a redesigned dynamic search box and AI-generated interactive layouts and widgets. It also introduces 'Search agents' specifically for real-time background monitoring (e.g., apartment listings and sneaker drops) and the merging of AI Overviews with AI Mode.

Agentic Evolution and Antigravity 2.0 Launch

The new source reveals an optimized version of Gemini 3.5 Flash that is 12x faster than competitors and details the launch of Antigravity 2.0, a standalone desktop IDE for agentic development. It also includes specific performance benchmarks against Gemini 3.1 Pro and quotes from DeepMind's chief technologist regarding the model's ability to build an operating system from scratch.