Google presents Gemini 3.5 Flash and Gemini Omni, two new pieces for its AI strategy in search, creation and agents

As expected, Google has taken advantage of its I/O 2026 event to announce several new features around Gemini. There are two especially notable names, namely, Gemini 3.5 Flash and Gemini Omni. The company positions the first as its new model focused on agentic tasks, programming and long-duration complex flows, while the second is presented as a multimodal proposal aimed at creating video from different types of input.

According to Google, Gemini 3.5 opens a new family of models and starts with Flash as the first generally available member. The company defines it as its strongest model to date for agents and coding, with the ability to execute development pipelines, manage research projects and work for several hours autonomously under supervision. The company has also confirmed that Gemini 3.5 Flash becomes the default model in the Gemini app and in AI Mode within Search globally.

Gemini 3.5 Flash, a fast model that surprises in synthetic tests

Google explains that 3.5 Flash has been designed to combine low latency with a level of performance closer to that of its frontier models. In its official data, the company maintains that outperforms Gemini 3.1 Pro in various programming benchmarks, agentive tasks and multimodal reasoning. Among the metrics cited are Terminal-Bench 2.1 with 76.2%, GDPval-AA with 1656 Elo, MCP Atlas with 83.6% and CharXiv Reasoning with 84.2%.

Another point that Google highlights is speed. The firm assures that Gemini 3.5 Flash is four times faster than other frontier models by measuring output tokens per second, and adds that it has developed an even faster optimized version for certain scenarios. That combination of speed and performance is the basis of its approach for agents, since it allows launching several sub-agents in parallel and sustaining long tasks over time.

The company links a good part of this strategy with Google Antigravityits agent-first development platform, which has also been updated during the event. Google showed on stage how multiple agents could share components of a project and then collaborate to build an operating system within that environment. Additionally, Flash integrates with Gemini API, Google AI Studio, Android Studio, Gemini Enterprise Agent Platform, and Gemini Enterprise.

Google has also detailed some uses already underway with partners. Examples cited include banks and fintech that automate multi-week workflows and data science teams that obtain insights in complex environments. In consumption, 3.5 Flash will be the basis of Gemini Spark, Google’s new personal agent that will work continuously to help manage the user’s digital life. The company will begin rolling it out to trusted testers and plans to bring the beta to Google AI Ultra subscribers in the United States next week.

In security, Google states that Gemini 3.5 has been developed under its Frontier Safety Framework and that reinforces safeguards against cyber and CBRN risks. The company also maintains that the model improves both in reducing harmful content and in reducing erroneous rejections of safe queries.

Gemini Omni is the blueprint for creating anything that takes physical laws into account.

Along with 3.5 Flash, Google has presented Gemini Omnia new model focused on multimodal creation that the company summarizes as a tool capable of “creating anything from any input,” starting with video. Its first variant, Gemini Omni Flash, begins to be deployed in the Gemini app, Google Flow and YouTube Shorts.

According to Google, Omni allows you to combine images, audio, video and text as input to generate videos, and then edit them through natural conversation, maintaining consistency in characters and elements. The company adds that the model improves the understanding of physical forces such as gravity, kinetic energy or fluid dynamics, and that it can also draw on Gemini’s knowledge of history, science and cultural context.

Google has also indicated that Omni can be used to create a digital avatar with the user’s voice and appearancealthough in this first phase the audio will be limited to voice references. The company notes that all videos will carry the SynthID digital watermark and that certain features, such as in-video audio and voice editing, remain in testing ahead of wider release. Gemini Omni Flash begins reaching Google AI Plus, Pro and Ultra subscribers, as well as YouTube Shorts and YouTube Create users starting this week.