google As the search giant races to keep up in model development and offer more agent services to its large user base, it’s rolling out the latest version of Gemini and new artificial intelligence models designed to simulate the physical world.
The company made the announcement Tuesday at the annual Google I/O developer conference, gaining an audience for new product debuts as the market focuses on the soaring valuations of OpenAI and Anthropic. Both companies are planning an IPO as early as this year.
At the heart of Google’s AI strategy is Gemini, its family of models and tools. CEO Sundar Pichai said the company is showcasing the Gemini 3.5 Flash, a lightweight addition to its suite that offers cutting-edge features at half, and in some cases nearly a third, the price of comparable Frontier models.
Speaking to reporters ahead of Tuesday’s event, Pichai said Gemini 3.5 Flash is “very fast.” The company said 3.5 Flash will be the default model for Gemini apps and AI mode in search worldwide.
“You no longer have to sacrifice latency for quality,” Google said in a blog post. The company said it has strengthened Gemini 3.5 Flash’s cybersecurity defenses so it is “less likely to generate harmful content or mistakenly refuse to answer safe queries.”
Google says its heavier version, Gemini 3.5 Pro, is being used internally but won’t be widely available for distribution until next month.
Sundar Pichai, Chief Executive Officer of Alphabet Inc., attends the Google I/O Developer Conference on Tuesday, May 19, 2026 in Mountain View, California, USA.
David Paul Morris | Bloomberg | Getty Images
On the agent AI front, Google announced Gemini Spark, a new general-purpose AI agent for Gemini apps that can infer information within connected apps. Google said it wants to help users navigate their digital lives by “acting on your behalf, under your direction.” Gemini Spark is in beta and will be available first to trusted testers and Google AI Ultra subscribers starting next week.
As more internet users gravitate toward chatbots, Google is trying to convince traditional search users that they can be trusted with tasks that require minimal input. As the company’s capital spending soars, Wall Street wants Google to show it can create deeper integrations between its products, and Agent could be a way to do that.
Expectations for AI companies continue to rise, especially considering Anthropic’s recently released Mythos model. The Mythos model is so powerful that it is said to have discovered thousands of previously unknown vulnerabilities in the world’s software infrastructure.
Google’s AI portfolio includes Omni, a world model designed to simulate the physical environment and predict what will happen next based on your actions. World models are commonly used in robotics and games, and have been thoroughly researched by DeepMind over the years.
Omni works with Flash, the Gemini app, Google Flow, and YouTube Shorts, and supports images and audio, the company said, adding in a separate blog post that users can let Omni edit their videos and create more realistic images.
“Take the video you took and ask Omni to change what is happening,” the post reads. The AI can “edit actions and add new characters and objects.”
Spotlight: What you can expect from Google I/O
