Back to the stories

Google DeepMind introduces Gemini Robotics 2.5

Score 9.2,

Autonomous AI Agents

A new AI model is designed to help robots learn a wider range of skills, faster, and work more reliably in less predictable surroundings.

Google DeepMind introduced Gemini Robotics 2.5 on November 10. It describes it as its most capable model yet for linking what a robot sees, what it is told, and how it acts. The model is rolling out to robotics partners.

That matters because robots need more than dexterity to work outside carefully arranged settings. They also need to act reliably and safely when their surroundings are less predictable.

Think of the model like a coordinator who takes in a scene and an instruction, then helps turn them into a movement. That is the basic idea behind AI that links vision, language and action.

For robotics partners, the next step is seeing whether the intended improvements hold up in real settings. A partner rollout is not proof that robots can safely handle any unfamiliar task.

The real test is whether broader, faster learning can make robots dependable in the messy situations people actually work in.