at the 2026 google i/o developer conference, gemini spark made its official debut—a truly “24/7 personal ai collaborator” that goes beyond simple q&a. it can understand natural language instructions, autonomously plan tasks, and execute multiple operations in parallel. running within an isolated, secure sandbox, it keeps tasks active in the background even when the device is asleep or the screen is off, without ever accessing local sensitive data.
this intelligent agent is powered by an upgraded gemini 3.5 model combined with google’s brand-new antigravity architecture, natively supporting deep integration with toolchains. in the future, it will open up a third-party plugin ecosystem via the mcp (model context protocol). during the demo, users simply need to say aloud, “help me organize my important emails from the past three days and plan a weekend gathering with friends,” and gemini spark will automatically invoke gmail, calendar, maps, and shopping platforms to complete the entire workflow, with results instantly synced to the notification center.
in its initial launch phase, gemini spark will be available starting next week to google ai ultra subscribers in the united states. the basic version costs $100 per month (approximately rmb 681), while the flagship edition is offered at a limited-time special price of $200 per month (regular price: $250). starting this summer, it will also be deeply integrated into the chrome browser, enabling one-click activation and seamless collaboration directly within web pages.
the industry widely regards this as a pivotal turning point in the evolution of ai assistants—shifting from “command‑driven responses” to “intention‑driven execution.” while ensuring privacy and stability, it redefines the boundaries of human‑machine collaboration efficiency.
