Google unveiled its next-generation lightweight artificial intelligence (AI) model along with an active agent built on it. It also introduced the biggest overhaul of its search system in 25 years and a next-generation video generation model, pushing to expand an AI ecosystem that spans everything from search to workflow automation and content creation.
On the 19th (local time), at its annual developer conference Google I/O held at Shoreline Amphitheatre in Mountain View, California, Google announced its next-generation model, Gemini 3.5 Flash.
Gemini 3.5 Flash is a lightweight model that nonetheless posted higher performance than the previous top model, Gemini 3.1 Pro, on most benchmarks, including agents, coding, and financial analysis.
In particular, it outperformed the top publicly available models of Gemini, GPT, and Claude on the agent rules Model Context Protocol (MCP) and on financial analysis benchmarks. In coding, it achieved scores close to GPT-5.5 on terminal environment benchmarks. However, on SWE-bench Pro, a general coding ability metric, it scored lower than Claude Opus 4.7 and GPT-5.5.
As a lightweight model that offers output speeds up to four times faster than other top-tier models, Gemini 3.5 Flash is expected to be highly useful in the workplace.
Chief Executive Officer Sundar Pichai said, "You may have heard anecdotes about corporations burning through their annual (AI) token budgets before May," adding, "If a company that uses 1 trillion tokens a day shifts 80% of its workload to models like Flash, it can save more than $1 billion a year."
Google also unveiled Gemini Spark, an active agent based on Gemini 3.5. The service supports summarizing a user's emails, preparing daily briefings, handling repetitive tasks, complex coding, and carrying out long-term projects. It runs 24 hours a day even when a laptop is closed or a smartphone screen is off.
Search, Google's core service, is getting its biggest change in 25 years. Users can attach not only text but also images, files, and videos to their searches, and visual aids and widgets that aid understanding are generated in real time on the results screen. An interface that smoothly transitions from the AI Overview at the top of search results to a conversational AI mode is also being introduced.
In shopping, Google will roll out Universal Cart in the United States this summer, which links Search, Gemini, YouTube, and Gmail to automatically handle everything from price tracking to payment.
Google also unveiled Gemini Omni, a multimodal video generation model that can process all forms of input and output, including text, voice, images, and video.
Whereas the existing video model VEO only converted text into video, Gemini Omni also features the ability to change characters in existing videos or transform styles. Google said it understands the laws of physics and can produce far more realistic videos.
Videos generated with Gemini Omni will carry a SynthID digital watermark that can identify whether they were AI-generated.
Gemini 3.5 Flash is available to all users starting today. Gemini Omni is offered to paid subscribers, and Gemini Spark will operate as a pilot service for Ultra plan subscribers.