Google DeepMind has unveiled Gemini Robotics 2, a new family of AI models built to control robots rather than chatbots.
According to the Economic Times, the models let robots move their entire bodies — not just an arm or a gripper — and handle delicate tasks. The report also describes a separate model in the family, ER 2, that plans and manages more complex work, suggesting a split between the system that thinks through a job and the system that physically executes it.
The headline number comes from Korean outlet 조선일보 (Chosun Ilbo), which reports that Gemini Robotics 2 achieves 92% precision on hand tasks. That figure is what most coverage has latched onto, though the sources here do not specify which tasks were tested or how the benchmark was constructed.
Investors noticed. TradingKey reports that the new robot model fueled a roughly 3% rally in Alphabet stock, while noting that the $360 level has remained hard for the shares to break.
Some context on why the precision claim carries weight: manipulation — picking up, positioning, and handling objects without dropping or crushing them — has long been the bottleneck in robotics. Language models got good at reasoning years before robots got good at hands. A model family aimed squarely at whole-body movement and delicate handling is an attempt to close that gap.
It matters because the distance between an AI that can describe a task and a machine that can actually perform it is the last thing standing between today's chatbots and robots that do useful physical work.