Overview
Gemini Robotics 2.5 is a vision-language-action (VLA) model for controlling robots in real-world environments. It converts visual and language inputs into motor control and is designed to improve the breadth, dexterity, reliability, and speed of robotic skills. Google announced it on November 10, 2026. (deepmind.google)
Capabilities and uses
- Supports different robot platforms, including Apptronik’s Apollo 2 humanoid and the dual-arm Franka Duo with parallel grippers.
- Google evaluated it across more than 200 tasks, including serving food with tongs, tying a bow, pipetting liquid, and manipulating delicate or deformable objects. Compared with previous baselines, Google reports 30%–100% higher task success rates, depending on the task, and 20%–30% faster completion. (deepmind.google)
- In a Mercedes-Benz production-line proof of concept, Apollo 2 used Gemini Robotics ER to track the workflow and Gemini Robotics 2.5 to control its hands and arms while sorting automotive parts. Google reports a 96% success rate for this pilot.
- Inference-time reasoning guidance steers robot actions around physical hazards. Google reports improvements in safe task completion of 33% around obstacles and 43% around nearby people in its evaluations. (deepmind.google)
Availability and limitations
Google says the model is rolling out to robotics partners. The reported performance and safety figures are evaluation or pilot results, not guarantees for all robots, tasks, or environments. The announcement does not establish general public API availability or specify an API model ID. (deepmind.google)
Pricing
Not verified: the official announcement does not specify a currency or input/output billing rates for this model.