Gemini Robotics 2.5 advances precise manipulation and multi-step robot tasks
Google DeepMind's new vision-language-action model reports higher success rates than Gemini Robotics 2 across precise manipulation, contact-rich tasks and reliability on multiple robot embodiments.
AT A GLANCE
Key points
- Provider evaluations report higher success rates on precise and contact-rich tasks.
- The same core model is demonstrated across different robot embodiments.
- The published results are DeepMind's own evaluations.
What is changing?
Google DeepMind has introduced Gemini Robotics 2.5, a vision-language-action model designed to translate visual instructions into physical robot actions. The company reports higher success rates than Gemini Robotics 2 on tabletop placement, precision insertion, tool use and multi-step contact-rich tasks. The same core approach is demonstrated on humanoid robots such as Apptronik Apollo and on dual-arm systems.
DeepMind's examples include packing pastries with tongs, untying ribbons, closing a toolbox and sweeping debris into a dustpan. The company says safety, reliability and adaptation to unfamiliar settings have improved. The reported gains, however, come from the provider's own evaluations, while the announcement offers limited detail about broad access.
Explore the original sources
Source date: 9 November 2026.
Published by AI Radar: 10 Nov 2026, 18:11 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…