11 demos of Gemini Omni and Gemini 3.5 in action

Source: Google Date: 2026-05-29 URL: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-3-5-videos/

Summary

Google’s I/O 2026 demo reel covers Gemini Omni (native video generation and multi-turn editing via natural language) and Gemini 3.5 Flash (positioned as the agentic workhorse: long-horizon task execution, workspace automation, and custom generative UI). Omni handles iterative video edits — repositioning subjects, changing environments, adjusting angles — while preserving character consistency. Gemini 3.5 Flash demos cover information agents, automated asset pipelines, personal dashboards, and Gmail/Docs/Slides orchestration.

Implications

This signal feeds the model layer and agentic engineering patterns threads. Google is shipping two distinct capability tiers in the same release window: Omni as a multimodal frontier play (direct competition with OpenAI’s video generation work) and 3.5 Flash as the cost-optimized agent backbone. The implications:

• The agentic-tier positioning of 3.5 Flash — explicitly framed as fast and capable for multi-step workflows — signals that Google intends Flash-class models to be the runtime for agent loops, not just the cheap inference option.
• Workspace integration demos (Gmail, Docs, Slides) are a competitive counter to Microsoft Copilot’s enterprise suite story; Google is demonstrating that agent orchestration works inside its own productivity stack.
• The native video generation in Omni accelerates the multimodal frontier race: Sora, Runway, and Kling now have a Google-native competitor with the distribution advantage of being embedded in Google’s toolchain.