How is Gemini Robotics 2 different from Gemini Robotics 1.5?

Gemini Robotics 2 moves beyond the upper-body table-top skills of Gemini Robotics 1.5. The new model controls the full humanoid — walking, crouching, stretching, and reaching — with multi-finger hands that can tie knots and seal ziplock bags. It also lets several robots plan and act together on one task.

Which robots does Gemini Robotics 2 support?

Google DeepMind names three launch partners for Gemini Robotics 2: Apptronik, Boston Dynamics, and Agile Robots. The model can adapt to a new bi-arm or humanoid embodiment in a few hours, typically with fewer than 200 example demonstrations, so more robot platforms can be added later.

Can developers try Gemini Robotics 2 today?

Gemini Robotics ER 2, the orchestration brain, is live on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform. The whole-body VLA and On-Device 2 models are behind a Trusted Tester waitlist, so most developers cannot deploy Gemini Robotics 2 on their own robots yet.

What tasks does Gemini Robotics 2 actually finish?

DeepMind reports Gemini Robotics 2 hits 45.7%–76.3% success on whole-body manipulation tasks and 74.2%–89.6% with parallel grippers, with multi-finger dexterity between 32% and 92% depending on the task. Demos include screwing in a light bulb, tying knots, folding clothes, and two robots handing objects across a shared workspace.

Google DeepMind · 2026-07-30 · major

Gemini Robotics 2 — DeepMind's new humanoid model controls the full body

Gemini Robotics 2 is Google DeepMind's next humanoid model. It controls a full humanoid body from feet to fingertips, coordinates multi-robot teams, and adapts to a new robot in a few hours with under 200 examples. Apptronik, Boston Dynamics, and Agile Robots are the launch partners.

Apptronik humanoid robot controlled by Gemini Robotics 2 reaching to manipulate objects on a table

DeepMind's new humanoid brain walks, crouches, and ties knots — and coordinates two robots at once.

Key specs

Whole body task success	45.7%–76.3%
Adaptation examples	<200
Partner robots	Apptronik · Boston Dynamics · Agile Robots

Quick facts

Maker	Google DeepMind
Models	Gemini Robotics 2 (VLA) + Gemini Robotics ER 2
What's new	Whole-body humanoid control, feet to fingertips
Dexterity	22-DoF hands, grippers, tying knots
Partner robots	Apptronik, Boston Dynamics, Agile Robots
Availability	ER 2 on AI Studio; VLA + On-Device via waitlist
Adaptation	New robot in hours with <200 examples

What is it?

Gemini Robotics 2 is a Vision-Language-Action model that steers a whole humanoid instead of only the arms. Google DeepMind pairs it with Gemini Robotics ER 2, a higher-level 'brain' that plans and calls tools. Partners for the launch are Apptronik, Boston Dynamics, and Agile Robots.

How does it work?

Whole-body control lets Gemini Robotics 2 send coordinated commands from feet to fingertips, driving 22-DoF hands or simpler grippers. ER 2 handles long-horizon planning, moment-finding on live video (91.3% accuracy), and hand-off between multiple robots. New embodiments are adapted with under 200 example demonstrations in a few hours.

Why does it matter?

Humanoid startups have had brains that can talk and arms that can grip, but rarely one model that can crouch, walk, reach, and dexterously manipulate through a single API. Gemini Robotics 2 pushes the frontier lab into full-body control at the same moment Apptronik Apollo and Boston Dynamics Atlas are hitting factory floors.

Who is it for?

Robotics teams and humanoid startups

Frequently asked questions

How is Gemini Robotics 2 different from Gemini Robotics 1.5?: Gemini Robotics 2 moves beyond the upper-body table-top skills of Gemini Robotics 1.5. The new model controls the full humanoid — walking, crouching, stretching, and reaching — with multi-finger hands that can tie knots and seal ziplock bags. It also lets several robots plan and act together on one task.
Which robots does Gemini Robotics 2 support?: Google DeepMind names three launch partners for Gemini Robotics 2: Apptronik, Boston Dynamics, and Agile Robots. The model can adapt to a new bi-arm or humanoid embodiment in a few hours, typically with fewer than 200 example demonstrations, so more robot platforms can be added later.
Can developers try Gemini Robotics 2 today?: Gemini Robotics ER 2, the orchestration brain, is live on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform. The whole-body VLA and On-Device 2 models are behind a Trusted Tester waitlist, so most developers cannot deploy Gemini Robotics 2 on their own robots yet.
What tasks does Gemini Robotics 2 actually finish?: DeepMind reports Gemini Robotics 2 hits 45.7%–76.3% success on whole-body manipulation tasks and 74.2%–89.6% with parallel grippers, with multi-finger dexterity between 32% and 92% depending on the task. Demos include screwing in a light bulb, tying knots, folding clothes, and two robots handing objects across a shared workspace.

Try it

Gemini Robotics ER 2 in Google AI Studio; whole-body VLA via the Trusted Tester waitlist.