Google DeepMind releases Gemini Robotics 2.5

Background

Gemini Robotics 2 is the earlier model in the same family. Google DeepMind compares it with Gemini Robotics 2.5 in a chart of success rates ^[1].

What happened

On 10 November 2026 Google DeepMind announced Gemini Robotics 2.5 ^[1]. The company called it its most capable vision-language-action model to date, and said it increases both the breadth of skills a robot can learn and the speed of learning ^[1]. Example tasks in the announcement include serving food with tongs, tying a bow on a gift box and pipetting liquid ^[1]. Pipetting means transferring small, measured amounts of liquid, as in a laboratory.

Key details

The chart compares Gemini Robotics 2 and 2.5 success rates on tasks such as closing a toolbox lid (0% to 93%) and packing pastries with tongs (5% to 100%) ^[1]. In a factory kitting pilot with Apptronik and Mercedes-Benz, the Apollo 2 humanoid reached a 96% success rate, per Google DeepMind ^[1].

Google DeepMind released the open-source ASIMOV-Jailbreak safety benchmark alongside the model ^[1].

Impact

The model is rolling out to robotics partners ^[1].