RELEASE·Nov 11, 2026, 1:58 PM ET

Google DeepMind introduces Gemini Robotics 2.5 with faster task execution

TL;DR:

  • Google DeepMind tested Gemini Robotics 2.5 on more than 200 different tasks.
  • Google DeepMind reported 96% success for automotive-parts packing in a Mercedes-Benz proof-of-concept test.
  • Google DeepMind said task completion was 20% to 30% faster than the previous baseline, with success rates ranging from 30% to 100%.
  • Google DeepMind is distributing the model through robotics partners including Agile Robots SE, Apptronik Inc. and Boston Dynamics Inc.

Summary

Google DeepMind introduced Gemini Robotics 2.5, a model for controlling robots across different hardware platforms. The company reported faster execution and a 96% success rate in a Mercedes-Benz packing trial, though broader task results ranged from 30% to 100%. Those differences matter when judging suitability for factory work. Separately, Google DeepMind announced AlphaGenome 2 for predicting how DNA sequences and variations affect human cell types.

Open siliconangle.com in a new tab ↗
Powered by Moatly Lens

The Bigger Picture

The headline tells you what happened. Discover the facts behind it, the forces at work, the connections that matter, and what comes next.

Why This Matters

This vision-language-action model proves that specialized factory automation is hitting its stride, even if general-purpose utility remains out of reach. While a 96% success rate at Mercedes-Benz shows real promise for repetitive logistics, the steep performance drop in other tasks highlights a significant reliability gap. Understanding this divide is vital for gauging how quickly these systems can move from controlled assembly lines into unpredictable human environments.

What We Actually Know

  • The new vision-language-action model controls diverse hardware, including humanoid robots with five-finger hands and platforms with two-fingered grippers.
  • In a real-world trial at Mercedes-Benz plants in Germany and Hungary, the system successfully sorted metal brackets with a 96% accuracy rate.
  • Task execution speeds have improved by up to 30% compared to earlier versions.

How the Pieces Connect

  • The system uses real-time reasoning to steer around hazards, addressing the safety needs of robots working in unpredictable human spaces.
  • While the Mercedes-Benz trial showed high reliability, performance across 200 other tasks varied significantly, dipping as low as 30%.

What This Could Change

  • InferenceThe wide gap in success rates suggests the model is ready for specific repetitive factory roles but might struggle with general-purpose utility.
  • ScenarioDiscrepancies in safety reporting could lead to confusion regarding how resilient the actual robot-control software is against malicious prompts.

Where the Story Goes Next

  • Which specific tasks resulted in the lower 30% success rate during testing?
  • Does the 99.5% jailbreak resistance score officially apply to the robotics model itself?