Google presented on Tuesday, during its annual conference at the Shoreline Amphitheatre, two new artificial intelligence models — Gemini 3.5 Flash and Gemini Omni — and an integrated virtual agent capable of operating autonomously 24 hours a day. The announcements mark a new step in the company's race to offer faster and multimodal AI solutions.
Gemini 3.5 Flash is a model designed for agentic coding and real-world workflows. According to Sundar Pichai, chief executive of Alphabet, Google's parent company, the model is “very capable and really comparable to the best models, but it is remarkably fast.” The company claims it outperforms Gemini 3.1 Pro on nearly all benchmarks and processes output tokens at a speed four times greater than direct competitors, reducing costs to less than half.
Internally, Google developers already use 3.5 Flash on the Antigravity 2.0 platform, which, according to Pichai, “dramatically accelerated how Google builds.” The 3.5 Pro version is expected next month.
In parallel, Gemini Omni was introduced, a natively multimodal model that accepts text, audio, images, and video as input and initially generates video outputs, with plans to add audio and image soon. Demis Hassabis, chief executive and co-founder of Google DeepMind, highlighted that the model is capable of “achieving a new level of world understanding, multimodality, and editing.” Hassabis mentioned that previous models such as Leo, Nano, Banana, and Genie already created videos and realistic simulations, but that Omni represents “a radical shift in simulating phenomena like kinetic energy and gravity.”
Another launch was Gemini Spark, a personal agent integrated into Workspace that operates 24 hours on virtual machines connected to the cloud. Spark can automate research projects, draft reports by cross-referencing data from multiple applications, and manage schedules autonomously, without needing the device to remain on. “Yes, you can use it while your computer is closed,” Pichai said.
Google also presented Daily Brief, an assistant that collects information from applications connected to Gemini and sends a morning summary to the user based on Calendar events and Gmail updates. The feature will be available to subscribers of Google AI Plus, Pro, and Ultra plans in the United States starting Tuesday.
Regarding pricing, Google announced that the Google AI Ultra plan, which previously cost $249.99 per month, now starts at $100 monthly, with a $200 option that includes access to the Project Genie model. The entry-level plan, Google AI Plus, remains at $7.99 per month.