Model positioning
Gemini Robotics is a vision-language-action (VLA) model built on Gemini 2.0 that uses physical actions as its output modality and is designed to directly control robots. It differs from Gemini Robotics-ER, released at the same time, which focuses on embodied reasoning. (deepmind.google)
Capabilities and uses
- Generalization and interaction: Can handle new objects, different instructions, and new environments; understand everyday instructions in multiple languages; and adjust its actions in response to changes in the environment or instructions.
- Dexterous manipulation: Official demonstrations showed multi-step tasks requiring fine manipulation, such as folding origami and packing snacks into a sealable bag.
- Robot adaptation: The model was trained primarily on data from the dual-arm ALOHA 2 platform and was also demonstrated controlling a dual-arm platform based on Franka robotic arms. Google states that the model can be specifically adapted for Apptronik's Apollo humanoid robot.
- Application areas: Aimed at robotics research and future robot-assisted tasks in homes and workplaces. (deepmind.google)
Limitations and safety
Official demonstrations and benchmark results do not guarantee that the model will be reliable across all robots, environments, or tasks. Actual deployment requires safety measures tailored to the robot itself, such as avoiding collisions, limiting contact forces, and maintaining dynamic stability. The arrangement for invited testers in the original announcement concerned Gemini Robotics-ER and cannot be used to conclude that an API for this model has been made publicly available. (deepmind.google)
Pricing
Official input and output prices for the Gemini Robotics model described in this entry have not been verified; neither the currency nor the billing units have been verified. The Gemini Robotics ER version listed on the official pricing page is a different model, so its prices cannot be directly applied. Date checked: 2026-11-19. (ai.google.dev)
Release date and official sources
- Release date: 2025-03-12. (deepmind.google)
- Official announcement: Gemini Robotics brings AI into the physical world
- Official pricing page: Gemini Developer API pricing