Google DeepMind family of robot models: a VLA, ER 2 for reasoning and planning, and On-Device 2 for offline use.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Gemini Robotics 2 is described by creators as a family of three models from Google DeepMind for whole-body robot control. It includes a vision-language-action (VLA) model, Gemini Robotics ER 2 for high-level reasoning and planning, and Gemini Robotics On-Device 2 for running offline on the robot.
Demos shown by reviewers include improved dexterity on tasks such as unscrewing a light bulb, sealing a Ziploc bag with grapes and tying a trash bag. The ER model acts as the robot's "brain" for complex multi-step tasks. DeepMind's site presents Gemini Robotics as an intelligence layer for any kind of robot, and lists a later "Gemini Robotics 2.5" update on dexterity.
According to one creator, the models can be tried in Google AI Studio. The provided official page text does not give pricing, licensing, exact release date or technical details, so those are left unknown. Which components are available to developers versus limited to trusted partners is not confirmed in the supplied material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
A family of three models for whole-body robot control: a vision-language-action model, ER 2 for high-level reasoning and planning, and On-Device 2 for offline use. It adds better dexterity, such as unscrewing a bulb or sealing a Ziploc bag. It can be tried in AI Studio.
A new model that acts as the robot's brain for complex tasks. Demos included putting grapes in a Ziploc bag, unscrewing a light bulb and tying a trash bag.
Closest neighbours by meaning, not just by category.
Research · Google
Google DeepMind research demo in which a Gemini-based robot learned to play cornhole from first-person videos and practice.
Research · Google DeepMind
Gemini-powered evolutionary coding agent from Google DeepMind for algorithm and mathematical discovery.
Model · Google DeepMind
Google DeepMind's lightweight, low-cost Gemini 3.5 model aimed at fast, high-volume tasks.
Model · Google
Google DeepMind's family of open-weights models built from Gemini technology, from phone-sized to laptop-class variants.
Model · Xiaomi
Open-weights Vision-Language-Action robot foundation model from Xiaomi, pre-trained on 100K+ hours of UMI trajectories.
Model · Google
Google's rumored fast, low-cost Gemini model, reportedly positioned below Pro; not confirmed by official pages.