What happened
On 27 October 2026 Google Cloud published a post by Google DeepMind Robotics and Google Cloud staff saying that Gemini Robotics ER 2 is now available in public preview. The post calls it a high-level reasoning model for robots across different physical forms, combining video understanding over time, long-horizon planning and tool orchestration.
Key details
All details come from the Google Cloud post, read on 29 October 2026. The performance figures are Google’s and we have not tested the model.
- Access: public preview in Model Garden and Agent Studio, with the model ID
gemini-robotics-er-2-preview. The post links to documentation and a pricing page but states no prices. - Reported figures: 91.3% accuracy when pinpointing the exact video frame of a critical event, with a mean absolute frame distance of 0.96 seconds, at “4x the execution speed”. Google also says the model reads more than ten categories of instruments and can handle multi-minute tasks involving hundreds of decisions.
- Customers quoted: Boston Dynamics, DeepReach, Arm and ASTAR IMRE.
- Wording to note: the post says the model is available “on the Gemini platform”; the products it names are Model Garden and Agent Studio.
Why it matters
The model is an orchestration layer for robots rather than a motor-control system: it interprets video, plans and calls tools. A public preview lets developers test those claims on their own footage and tasks, which is the evidence that matters, since the published numbers come from Google’s own evaluations.
Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article. Details are from Google’s 27 October 2026 post, read on 29 October 2026.
Sources
- Gemini Robotics Embodied Reasoning 2 is now available on Google Cloud Primary source , Google Cloud Blog (27 October 2026)