Skip to content
Research · Nov 10, 2026

Google DeepMind unveils Gemini Robotics 2.5, a vision-language-action model for robotics

The model expands task breadth, improves reliability with a 96% success rate in a real-world automotive proof-of-concept, and introduces a new safety benchmark for adversarial robotics scenarios.

Trust79
HypeLow hype

1 source · cross-referenced

ShareXLinkedInEmail
TL;DR
  • Google DeepMind released Gemini Robotics 2.5, a vision-language-action model designed to enhance robot dexterity, reliability, and safety in real-world environments.
  • The model was evaluated across more than 200 distinct tasks, achieving a 30% to 100% higher success rate and completing tasks 20% to 30% faster than previous baselines.
  • In a real-world proof-of-concept with Apptronik’s Apollo 2 humanoid and Mercedes-Benz, the model achieved a 96% success rate in automotive kitting tasks.
  • Gemini Robotics 2.5 introduces a new safety benchmark to test against adversarial jailbreaks specific to robotics.
  • The model is being rolled out to robotics partners and supports multiple robot types, including Apptronik’s Apollo 2 and the Franka Duo platform.

Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action (VLA) model described as its most capable to date for robotics. The model is designed to enhance robot dexterity, reliability, and safety, enabling robots to operate in unstructured, real-world environments beyond the lab.

The model was evaluated across more than 200 distinct tasks, including serving food with tongs, tying a bow on a gift box, and pipetting a liquid. These tasks reflect a move toward human-level dexterity by combining advanced tool use, manipulation of deformable objects, and handling delicate items with precision. Compared to previous baselines, Gemini Robotics 2.5 achieved a 30% to 100% higher success rate depending on the task type, while completing tasks 20% to 30% faster.

Gemini Robotics 2.5 supports multiple robot types, including Apptronik’s Apollo 2 humanoid robot and the Franka Duo platform. The Apollo 2 features a five-finger, 22 degree-of-freedom SharpaWave hand, while the Franka Duo uses two-fingered parallel grippers.

In a real-world proof-of-concept, Google DeepMind partnered with Apptronik and Mercedes-Benz to test the model on active automotive production lines in Kecskemét, Hungary, and Sindelfingen, Germany. The Apollo 2 humanoid autonomously performed automotive kitting—sorting individual parts from bulk bins into organized trays in the exact order needed for each vehicle. The model achieved a 96% success rate in this task, demonstrating its reliability in an active workplace.

To address safety concerns as robots move into everyday environments, Google DeepMind introduced a new benchmark to test against adversarial jailbreaks specific to robotics. The company also emphasized improvements in physical safety steerability, aiming to ensure safer interactions in dynamic settings.

Gemini Robotics 2.5 is now rolling out to robotics partners, marking a step toward broader deployment of advanced robotics models in industrial and real-world applications.

Sources
  1. 01Google DeepMind — Blog — Gemini Robotics 2.5 accelerates real-world general dexterity
Also on Research

Stories may contain errors. Dispatch is assembled with AI assistance and curated by human editors; despite the trust-score filter, mistakes happen. We correct publicly — every article links to its revision history. Nothing here is financial, legal, or medical advice. Verify before relying on any claim.

© 2026 Dispatch. No ads. No sponsorships. No paid placement. Reader-supported via Ko-fi.

Built by a person who cares about honest AI news.