Google has opened public preview for Gemini Robotics ER 2, the embodied-reasoning line aimed at robotics rather than chat.
What shipped
Two endpoints are available. gemini-robotics-er-2-preview handles advanced spatial reasoning and multi-robot coordination — planning across several machines working in the same space rather than driving one in isolation. gemini-robotics-er-2-streaming-preview is optimised for real-time text streaming through the Live API, for cases where a robot cannot wait for a complete response before acting.
Migration deadline
The previous generation, gemini-robotics-er-1.6-preview, will be shut down on August 31, 2026. Anyone building on 1.6 has a hard date to move by. Preview endpoints carry no stability guarantee, and this is a reminder of what that means in practice.
Why it matters
Multi-robot coordination is the interesting part. Single-arm manipulation is a solved-enough problem that the frontier has moved to machines that share a workspace and have to reason about each other. That is the difference between a demo and a warehouse.
The streaming endpoint points at the same shift. Robotics has a latency budget that chat does not — a response that arrives after the moment has passed is worse than no response. Building a dedicated streaming path suggests the constraint is being taken seriously rather than papered over.
This remains preview software with a deprecation already scheduled for its predecessor. Useful for evaluation, not something to build a production line on yet.
Reading the deprecation
The shutdown of ER 1.6 on August 31 arriving on the same day as ER 2's preview is worth noting on its own. Roughly a month of overlap is a short migration window for robotics work, where changes have to be validated against physical hardware rather than a test suite.
Anyone evaluating this line should assume the same cadence applies going forward and budget for it. Preview endpoints in robotics carry a cost that preview endpoints in chat do not: the integration work sits partly in the physical setup, and that does not port automatically.
Where it fits
For most of our audience this is context rather than a tool to adopt today. It matters because embodied reasoning is where the large labs are spending next, and the capabilities land in adjacent products — spatial reasoning improvements show up in video understanding and screen control well before they show up in a warehouse.