Google DeepMind released Gemini Robotics 2, a set of three models that let robots interpret spoken instructions and control their whole body, including walking and balancing, rather than just their arms. It also adds finer hand control, longer multi-step tasks and the ability for several robots to work together. Only the reasoning model is broadly available; the models that drive movement are limited to early-access partners, and fine multi-finger tasks still often fail.
What changed
Earlier Gemini Robotics models controlled only a humanoid's upper body for table-top tasks and could not coordinate several robots.
What it unlocks
Directing a full humanoid robot to walk, bend and carry an object across a room from a spoken instruction, and having two robots split a task between them.
- Pick up from table 68.4%, floor 45.7%, shelf 76.3% (Apollo with Inspire hands)
- Multi-finger tasks: unscrew bulb 92%, screw bulb 36%, dustpan 32%
- Gripper tasks: precise insertion 89.6%, tool kitting 78.9%
- Adaptation to a new two-arm robot in a few hours with fewer than 200 examples
What you need to act on it
- Google AI Studio access for the reasoning model
- private preview on Gemini Enterprise Agent Platform
- early-access partner approval for the action and on-device models
- compatible robot hardware
- deepmind.google2026-07-30