Google DeepMind's Gemini Robotics 2.5 packs car parts with 96% success in a Mercedes-Benz trial
The new model lets robots see, follow instructions and act, and is 20–30% faster than the previous version.
- Tasks tested: 200+
- Speed gain: 20–30%
- Mercedes-Benz packing success: 96%
What's new
On 11 November Google DeepMind released Gemini Robotics 2.5, a model that controls robots using vision and language, built on its Gemini 4 Argon model. DeepMind says it was tested on more than 200 tasks and finishes them 20–30% faster than before, with success rates from 30% to 100% depending on the task. In a real-world trial at Mercedes-Benz, it packed automotive parts with 96% success.
Who gets it
It works on different robot bodies, including Apptronik's Apollo 2 humanoid, and is rolling out through partners such as Apptronik, Agile Robots and Boston Dynamics. DeepMind also announced AlphaGenome 2, which predicts how DNA changes affect gene activity in specific cell types; it is a research tool, not approved for clinical use.
