Google ships Gemini 3.5 Flash, Omni video model, and Spark agent at I/O 2026

Google's annual developer conference this week centered on speed and multimodality, with the company shipping three distinct Gemini models and a redesigned interface that folds video and voice into a single conversational thread.

Gemini 3.5 Flash is the headline language model. Google says it runs four times faster than Gemini 3.1 Pro while handling multi-step tasks, planning, report writing, and code generation. The model is now the default in the Gemini web interface and in AI Mode within Google Search. No parameter count or context length was disclosed in the announcement.

Gemini Omni is a multimodal video model that generates and edits video from text, image, or audio prompts. Google describes the output as "cinematic" and says the model supports conversational editing — users can refine clips through follow-up prompts. A lighter variant, Omni Flash, is already live in Gemini, Google Flow, and YouTube Shorts.

Gemini Spark is a background agent built on 3.5 Flash. It runs across Workspace and other Google services, searches local files, and assembles reports and task lists without active user input. Spark is available through Google Cloud.

The Gemini interface itself got a redesign under the name Neural Expressive. The update adds animations, new typography, and expanded formatting for answers — including inline graphics, timelines, and video embeds. Gemini Live, the voice mode, is now integrated into the main chat interface so users can switch between text and voice without losing context.

Google Search also picked up generative features: a dynamic search bar that adapts to the query type, support for text, image, video, file, and Chrome tab inputs, and mini-app widgets that render visual answers inline. The company is also rolling out SynthID image verification, which flags whether an image was AI-generated.