SYLIN

Gemini Robotics

by Google · Model · Robotics

unproven

Impressive claims from Google, but closed and untested by anyone outside.

VLA evaluated on 200 tasks: 30–100% higher success and 20–30% faster than prior baselines; rolling out to partners

AccessClosed

Why we say so

Google and DeepMind report 30-100% higher success than prior baselines across 200+ tasks, and 96% accuracy sorting car parts in a Mercedes-Benz proof of concept. These are all the maker's own figures, and the model is closed and rolling out only to partners, so independent testers have had no access.

Versions

VersionReleasedRank
Gemini Robotics 2.5LATEST
unproven Maker-reported gains only; partner access, no independent results
Nov 11
Gemini Robotics 2Aug 2

Evidence

  1. Nov 11
    Tested on 200+ tasks; 96% accuracy sorting car parts in an Apptronik and Mercedes-Benz proof of concept
  2. Aug 2
    Gemini Robotics 2 is a vision-language-action model for whole-body control, walking to fingertip dexterity; try it in AI Studio

Alternatives

GPT AstraCritic says the test is one small experiment, uses an IK pipeline, and skips comparison with VLA or world-action models such as Gemini Robotics