
Gemini Robotics 2
VLA + embodied reasoning + on-device model family for whole-body robot control.
About
Gemini Robotics 2, released July 30, 2026 by Google DeepMind, is a three-model physical-AI family: a vision-language-action (VLA) model for whole-body control, Gemini Robotics ER 2 for embodied reasoning and long-horizon planning, and Robotics On-Device 2 for low-latency edge deployment. The VLA treats the robot as one integrated system — legs, torso, arms, hands — rather than independent joints, demonstrated on the Apptronik Apollo 2 humanoid, Boston Dynamics Spot, and Franka arms. DeepMind also released ASIMOV-Agentic, the first dedicated safety benchmark for autonomous robot operation.
The honest picture: access is limited to trusted testers, robotics partners, and a small group of OEMs — this is not a model you can download and try. DeepMind's own robotics director has said true dexterity remains distant, with robots still slow and deliberate, and multi-robot collaboration demos are lab-grade, not production deployments.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 57; Trending score 88/100 from researcher assessment.