Google DeepMind just gave robots a serious upgrade — and this one is already working a real job.
On November 11, DeepMind debuted Gemini Robotics 2.5, its most capable vision-language-action model yet. Think of it as the robot's brain: it sees the world, understands instructions in plain language, and turns that into physical action. This version is built on the reasoning power of Gemini 4 Argon, Google's flagship model, which DeepMind says lets robots learn new skills faster and apply them to real-world situations.
What Got Better
So what actually got better? Three things, according to DeepMind. Dexterity is up — robots can handle delicate, flexible objects and use everyday tools with more accuracy. Reliability is up — in a real-world proof of concept at a Mercedes-Benz plant, robots running the model packed automotive parts with a 96% success rate. And safety is up too, with improved physical steerability and a new benchmark for testing the model against jailbreaks.
The Numbers
The numbers back it up. Across more than 200 tasks, Robotics 2.5 completes jobs 20 to 30 percent faster than its predecessor, with success rates ranging from 30 to 100 percent depending on how hard the task is.
The model works across different robot bodies, including Apptronik's Apollo 2 humanoid and the Franka Duo platform, and it is rolling out through DeepMind's robotics partners, including Agile Robots, Apptronik, and Boston Dynamics. Apptronik's CEO Jeff Cardenas said the models improved dramatically in just a few weeks of post-training and carried over to Mercedes-Benz production with very little drop-off.
Why It Matters
Why it matters: most robot demos still happen in carefully controlled labs. A 96% success rate packing real car parts on a real factory floor is a different kind of proof — it is the moment robot AI stops being a demo and starts being a coworker. The same week, DeepMind also announced AlphaGenome 2, a model that predicts how DNA variants affect individual cell types — a reminder that the lab is shipping on every front at once.
The robot workforce is not coming someday. In Stuttgart, it is already clocking in.
Comments
Post a Comment