Back to news
launchGoogle2026-11-11

Google unveils Gemini Robotics 2.5, reaching a 96 percent success rate for parts picking and packing on a Mercedes-Benz line

Google released Gemini Robotics 2.5, its most capable vision-language-action model to date, tested across more than 200 tasks with success rates 30 to 100 percent above the previous baseline and a 96 percent success rate picking and packing automotive parts on a Mercedes-Benz line.

Google released Gemini Robotics 2.5 on November 10, described as its most capable vision-language-action model to date, substantially expanding the breadth of skills robots can learn and the speed at which they learn them, while improving dexterity, reliability and safety so robots can operate in less structured real environments. Google attributed much of the progress to the ability of Gemini 4 Argon to accelerate robotics development. The model was tested across more than 200 different tasks covering task breadth, performance and complexity, including dispensing food with grippers, tying a bow on a gift box and pipetting liquids, tasks that combine advanced tool use, complex deformable object manipulation and delicate handling of fragile items. Evaluation showed success rates 30 to 100 percent above the previous baseline depending on task type, and task completion speeds 20 to 30 percent faster. The model natively supports multiple robot types including the Apptronik Apollo 2 humanoid with a five-finger 22-degree-of-freedom SharpaWave hand and the Franka Duo platform with two-finger parallel grippers. Working with Apptronik and Mercedes-Benz, Google deployed Apollo 2 on in-use automotive production lines in Kecskemet, Hungary and Sindelfingen, Germany, where the robot autonomously picked metal car brackets from bins, placed them into sorting containers and corrected assembly gaps throughout, reaching a 96 percent success rate. On safety, the model uses inference-time reasoning guidance to avoid physical hazards in real time, raising safe task completion by 33 percent around obstacles and 43 percent around people. Google also published the open-source ASIMOV-Jailbreak benchmark, on which Gemini 4 Argon achieves a 99.5 percent jailbreak resistance rate versus 82 percent for the previous Gemini Robotics ER 2.

Gemini Robotics 2.5GoogleVLA机器人具身智能自动驾驶