Google Unveils Gemini Omni Video Generation and Gemini 3.5 Flash for Agentic AI
Google announced Gemini Omni, a multimodal model that generates and edits video through natural language, and Gemini 3.5 Flash, optimized for complex agent workflows.
Last verified:
Gemini Omni and 3.5 Flash Announced at Google I/O 2026
Google announced two new model releases at its 2026 I/O developer conference: Gemini Omni, a video-generation model that accepts multimodal input and supports conversational editing, and Gemini 3.5 Flash, a reasoning model designed for autonomous agent workflows. According to the Google AI Blog, Gemini Omni bridges creation and reasoning by accepting images, audio, video, and text to generate and iteratively refine video content, while Gemini 3.5 Flash prioritizes agentic execution over creation, targeting complex long-horizon tasks with real-world utility.
Gemini Omni’s Natural-Language Video Editing Capability
The defining feature of Gemini Omni is its ability to edit video through conversational prompts rather than traditional video editing tools. According to Google AI Blog, users can provide multi-turn instructions that build sequentially, with the model maintaining character consistency, physics accuracy, and scene context across edits. Demonstrated examples include transforming a sculpture into bubbles, dimming lights, creating recursive visual effects, and repositioning subjects across environments—all without requiring re-recording or manual frame-by-frame work.
The model’s iterative refinement capability allows users to modify environments, camera angles, visual styles, and specific objects while preserving the original scene’s foundational logic. Google’s demonstration showed a violinist being repositioned across four editing turns: initial video generation, environment transport, object removal, and camera angle adjustment, with each prompt building on the previous result.
Gemini 3.5 Flash for Agentic Workflows
While Gemini Omni addresses multimodal content creation, Gemini 3.5 Flash represents a shift toward autonomous agent capabilities. According to the Google AI Blog, Gemini 3.5 Flash combines frontier-level intelligence with action-execution capacity, excelling at complex long-horizon tasks and coding applications. The model is positioned as the first in a broader Gemini 3.5 family, suggesting multiple variants optimized for different operational scales and complexity profiles.
The emphasis on “frontier performance for agents” indicates Google is positioning Gemini 3.5 Flash as an alternative to dedicated reasoning-plus-action architectures, potentially competing with OpenAI’s reasoning models and Anthropic’s agentic Claude variants in environments requiring chained multi-step execution.
Why This Matters
The dual release of Gemini Omni and Gemini 3.5 Flash reflects a deliberate product segmentation: Omni targets creative professionals and content creators requiring sophisticated video synthesis and editing automation, while Gemini 3.5 Flash targets enterprise teams building autonomous systems and software-engineering agents. The natural-language video editing capability in Omni lowers barriers for non-technical creators, potentially expanding the addressable market for AI-native content tools. For agentic systems, Gemini 3.5 Flash’s positioning as a reasoning-plus-action model in a frontier-performance tier signals Google’s commitment to the agent-building market, where latency, reasoning depth, and execution reliability directly impact production adoption. Developers evaluating agent architectures will likely include Gemini 3.5 Flash in benchmarking against existing reasoning-first models.
Frequently Asked Questions
What is Gemini Omni and what inputs does it accept?
According to Google AI Blog, Gemini Omni is a multimodal generative model that accepts images, audio, video, and text as input to generate high-quality videos grounded in real-world knowledge. It also supports iterative natural-language video editing where each instruction builds on prior context.
How does Gemini Omni handle video editing through conversation?
Gemini Omni allows users to edit videos via natural-language prompts. The model maintains character consistency, physics accuracy, and scene memory across multiple editing turns, enabling both targeted modifications and complete scene transformations without re-recording.
What is Gemini 3.5 Flash and what is it optimized for?
Gemini 3.5 Flash is the first release in Google's latest model family, designed to combine frontier-level reasoning with agentic execution. It is optimized for complex long-horizon tasks, agent-based workflows, and coding applications.
What distinguishes Gemini 3.5 from Gemini Omni?
Gemini Omni focuses on multimodal video generation and iterative editing. Gemini 3.5 Flash is architected for autonomous agent behavior and complex sequential reasoning tasks that require real-world action.