Overview
Gemini Robotics 1.5 is a vision-language-action (VLA) model for controlling robots. It converts visual information and instructions into motor commands for physical tasks. Google DeepMind announced it on September 25, 2025. (deepmind.google)
Capabilities and uses
- Reasons about actions in natural language before executing them and can divide longer tasks into shorter steps.
- Learns across robot embodiments. Google demonstrated transfer of tasks among ALOHA 2, a bi-arm Franka platform, and Apptronik’s Apollo humanoid.
- Supports multi-step manipulation tasks, such as sorting laundry. It can work with the separate Gemini Robotics-ER 1.5 model, which provides high-level planning and instructions while the VLA model performs actions. (deepmind.google)
Access and limitations
At launch, Google said Gemini Robotics 1.5 was available to select partners; current access terms have not been verified. The Gemini API availability described in the announcement applied to Gemini Robotics-ER 1.5, a separate vision-language model, not to this VLA model. Google’s demonstrations do not establish reliable performance for every robot, task, or environment. The model’s safety approach includes reasoning before acting and triggering robot-side safety subsystems when needed; deployment still requires appropriate physical safety measures. (deepmind.google)
Pricing
As of November 18, 2026, an official input price, output price, billing unit, and currency for Gemini Robotics 1.5 could not be verified. Pricing for other Gemini Robotics models should not be assumed to apply. (ai.google.dev)
Official sources
- Google DeepMind announcement (deepmind.google)
- Google Gemini API pricing — checked November 18, 2026. (ai.google.dev)