Everything Google announced at I/O 2026, including Gemini Omni, new Spark AI agent and Gemini 3.5 Flash

Google showcased some of its most exciting AI updates at I/O 2026 at the Shoreline Amphitheatre in Mountain View, California. During the keynote address headlined by CEO Sundar Pichai, the company unveiled its latest Gemini 3.5 Flash language model, its agentic AI approach with Gemini Spark, a new ‘world model’ series with Gemini Omni, and a refreshed coding assistant in Antigravity 2.0.

Here are the biggest updates you should know from I/O 2026:

1) Gemini Omni:⁠⁠

Google has introduced a new world model called Gemini Omni. The new model combines the reasoning abilities of other generative media models from Google, including Veo, Nano Banana, and Genie.

While on stage at I/O 2026, Google DeepMind CEO Demis Hassabis called Omni a significant leap toward artificial general intelligence (AGI).

Unlike traditional AI video generation tools that rely mainly on prompts, Gemini Omni can combine text, images, videos, and audio references to generate high-quality videos while using Gemini's real-world understanding. The model can also be used to further edit generated videos using natural language prompts.

Google says Omni also has a stronger grasp of concepts like gravity, kinetic energy, and fluid dynamics, allowing it to generate more realistic visuals.

The first model under the new Gemini Omni family, Gemini Omni Flash, is rolling out starting today to Google AI Plus, Pro, and Ultra subscribers via the Gemini app and Google Flow. Google is also bringing Omni Flash to YouTube Shorts and YouTube Create.

2) Gemini 3.5 Flash

Google unveiled its latest Gemini 3.5 Flash model, which will now replace Gemini 3.1 Flash as the default model inside the Gemini app.

The tech giant claims its new model is four times faster than competing frontier AI models, allowing users to get responses significantly quicker without sacrificing capability.

On benchmarks, Google says Gemini 3.5 Flash outperforms the older Gemini 3.1 Pro model across several coding and agentic tests. Meanwhile, the model also manages to beat GPT-5.5 and Claude Opus 4.7 in multi-step tool use (MCP Atlas), financial analysis (Finance Agent v2), and complex visual comprehension benchmarks such as MMMU-Pro and CharXiv.

3) Google Search gets its biggest AI makeover

After adding AI Mode to Search at last year's I/O, Google is making major changes to how users search the web this time around. First up, AI Mode is now powered by Gemini 3.5 Flash, which the company announced at I/O 2026.

There is also a new customised Search box that expands dynamically to take longer inputs while also supporting multimodal input formats like text, images, files, and videos simultaneously.

Google is also updating autocomplete beyond simply predicting the next word. The feature can now offer AI-powered suggestions designed to help users formulate more complex questions by adding context and nuances they may not have considered.

However, the biggest upgrade is the introduction of Search Agents, which can monitor topics on behalf of users in the background. Google demonstrated examples of users creating agents to track apartment listings, financial updates, shopping information, and more.

Search can also be used to build tools, trackers, and dashboards, essentially "mini-apps", for long-running projects like planning a wedding or moving to a different city.

4) Gemini Spark:

Google also unveiled a new AI-powered agent experience called Gemini Spark. The company says its new "personal AI agent" is built to help users manage their digital lives while operating under their direction and control.

Google says the new agentic tool is deeply integrated with Workspace apps such as Gmail, Docs, Slides, and more, allowing it to take on more complex tasks across a user's digital life.

Unlike a traditional chatbot, Gemini Spark runs on dedicated virtual machines inside Google Cloud and is designed to continue working even after users close their laptops or switch off their devices.

Users can also create recurring tasks and triggers using Spark, such as scanning monthly credit card statements for hidden subscription charges or tracking school updates and deadlines before compiling daily summaries for parents.

5) Antigravity 2.0:

Google is not in the mood to lose the AI-powered coding race to OpenAI's Codex or Anthropic's Claude Code. As a result, the company introduced Antigravity 2.0 at I/O 2026 powered by Gemini 3.5 Flash, which it says can allow developers to go from an idea to a working application without writing code themselves.

Google announced a new Antigravity 2.0 desktop application, describing it as an "agent-first" workspace. The app acts as a central hub where users can manage projects and run multiple AI agents simultaneously.

To showcase the power of its new platform, Google ran an experiment where Antigravity was used to build a fully functioning operating system from scratch.

Google said AI agents wrote, tested, and audited every part of the operating system, from the scheduler and memory management to the file system itself. During the live demo, Google even used Antigravity to identify missing keyboard and video drivers before patching the OS and successfully running Doom on it.