Google DeepMind has introduced Gemini Robotics 2.5, its most capable vision-language-action model to date, and the headline result comes from a working car factory. In a proof of concept on active Mercedes-Benz production lines, Apptronik Apollo 2 humanoid robots running the model sorted automotive parts with a 96% success rate.

The announcement was published on 10 November 2026. DeepMind describes Gemini Robotics 2.5 as a model that connects what a robot sees and is told with the physical actions it takes, and says it increases both the range of skills a robot can learn and the speed at which it acquires them.

Humanoids doing kitting on live production lines

The factory test took place at Mercedes-Benz plants in Kecskemet, Hungary, and Sindelfingen, Germany. There, Apollo 2 robots carried out automotive kitting, the repetitive task of sorting individual metal brackets from bulk bins into sequencing trays in the exact order each vehicle on the line needs them.

Two models shared the work. Gemini Robotics ER, an embodied reasoning model, tracked the workflow, while Gemini Robotics 2.5 controlled the robot's hands and arms, picking brackets from supply bins, placing them into containers and correcting gaps. DeepMind says the pilot shows promising progress towards wider real-world industrial use, although neither company has said whether the humanoids will stay on the lines permanently.

Dexterity gains and a new safety benchmark

DeepMind evaluated the model across more than 200 distinct tasks, including serving food with tongs, tying a bow on a gift box and pipetting liquid. Compared with previous baselines, it reports success rates 30% to 100% higher depending on the task type, with tasks completed 20% to 30% faster. The model works with several robot types, including Apollo 2 with its five-finger SharpaWave hand and the Franka Duo platform.

Safety is part of the release too. DeepMind has published ASIMOV-Jailbreak, an open-source benchmark that tests whether robots can be talked into unsafe actions through disguised prompts. Its Gemini 4 Argon model achieved 99.5% jailbreak resistance in the company's evaluations, against 82% for the previous baseline. Gemini Robotics 2.5 is rolling out to DeepMind's robotics partners.

Source: Google DeepMind