Introduced Gemini Robotics 2.5, a vision-language-action model that turns camera input and instructions into robot movements, initially for robotics partners. (Nov. 10)ใ€2ใ€‘

On Nov. 10, 2026, Google DeepMind introduced Gemini Robotics 2.5, its latest vision-language-action model, which converts camera views and instructions into robot actions; it is rolling out to robotics partners, not the general public. Tested on more than 200 tasks, including tying bows, pipetting and serving food with tongs, it achieved success rates 30%โ€“100% above earlier baselines and completed tasks 20%โ€“30% faster, DeepMind reported. In a pilot with Apptronik and Mercedes-Benz, Apollo 2 humanoids sorted metal brackets on production lines in Hungary and Germany, reaching 96% success. The model supports different robot forms, from arms to humanoids, including Apolloโ€™s five-finger, 22-degree-of-freedom hand. DeepMind says real-time reasoning improves safe task completion by 33% around obstacles and 43% near people; it also released ASIMOV-Jailbreak, a benchmark of hundreds of disguised unsafe prompts.