← Back to GPN · All articles

Google DeepMind's Gemini Robotics 2.5 Makes Robots Faster Learners

OFFICIAL Written by GPN · Nov 10, 2026, 2:00 PM ET · · Tech

Google DeepMind has introduced Gemini Robotics 2.5, its most capable vision-language-action model to date. Announced on November 9, the model increases both the breadth of skills a robot can learn and the speed at which it can acquire them, pushing dexterity, reliability, and safety further toward real-world use.

DeepMind evaluated the model across more than 200 distinct tasks, from serving food with tongs and tying a bow on a gift box to pipetting a liquid. Compared with previous baselines, Gemini Robotics 2.5 achieves a 30% to 100% higher success rate depending on the type of task, while completing tasks 20% to 30% faster.

The real-world test came on Mercedes-Benz production lines in Kecskemet, Hungary, and Sindelfingen, Germany, where Apptronik's Apollo 2 humanoid robots performed automotive kitting: sorting individual parts from bulk bins into organized trays in the exact order needed for each vehicle on the line. The proof-of-concept hit a 96% success rate.

Safety work includes inference-time reasoning guidance that steers around physical hazards in real time, boosting safe task completion by 33% around obstacles and 43% around nearby people. DeepMind is also releasing ASIMOV-Jailbreak, a new open-source benchmark for testing jailbreak robustness in robotics, on which Gemini 4 Argon achieves 99.5% jailbreak resistance, up from 82% for the previous baseline.

Gemini 4 Argon also accelerated the model's development, acting as a simulation engine that can generate thousands of varied practice runs from a real-world camera scene and build a realistic digital twin. Gemini Robotics 2.5 is rolling out to DeepMind's robotics partners, and works with many robot types, including Apptronik's Apollo 2 and the Franka Duo platform.

Facts gathered from

• Google DeepMind

More on this

Share