---
type: "post"
id: "gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5"
slug: "gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5"
title: "Gemini Robotics ER 2 overview: Gemini Enterprise Agent Platform"
description: "Google Cloud’s documentation describes Gemini Robotics ER 2, a preview vision-language model for reasoning about physical environments and coordinating robot tasks."
retrieval_nugget: "Google Cloud’s documentation describes Gemini Robotics ER 2, a preview vision-language model for reasoning about physical environments and coordinating robot tasks. Its combination of spatial reasoning, video understanding, code execution, and custom robot APIs makes this a concrete robotics capability release, though not a general-purpose agent update."
published_at: "2026-11-03"
updated_at: "2026-11-07"
record_date: "2026-11-03"
date_kind: "published_at"
topics: ["gemini","google"]
entities: ["Google Cloud","Gemini Robotics ER 2","Gemini 3.5 Flash","Google Gen AI SDK","Gemini Enterprise Agent Platform"]
related_patterns: ["harness-architecture-outlives-model-choice"]
novelty_label: "same_event"
prior_refs: ["https://newruntime.com/posts/gemini-robotics-embodied-reasoning-2-enters-public-preview-e0fdff/"]
source_url: "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/gemini-robotics-er"
source_urls: ["https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/gemini-robotics-er"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5/","markdown":"https://newruntime.com/posts/gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5.md","json":"https://newruntime.com/posts/gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5.json"}
markdown_url: "https://newruntime.com/posts/gemini-robotics-er-2-overview-gemini-enterprise-agent-7314c5.md"
---

# Gemini Robotics ER 2 overview: Gemini Enterprise Agent Platform

## Retrieval answer

Google Cloud’s documentation describes Gemini Robotics ER 2, a preview vision-language model for reasoning about physical environments and coordinating robot tasks. Its combination of spatial reasoning, video understanding, code execution, and custom robot APIs makes this a concrete robotics capability release, though not a general-purpose agent update.

## Summary
Google Cloud’s documentation describes Gemini Robotics ER 2, a preview vision-language model for reasoning about physical environments and coordinating robot tasks. Its combination of spatial reasoning, video understanding, code execution, and custom robot APIs makes this a concrete robotics capability release, though not a general-purpose agent update.

## What happened
Google Cloud describes Gemini Robotics ER as a vision-language model that interprets visual data, reasons about space and time, plans multi-step tasks, and orchestrates robots and tools. The newly documented Gemini Robotics ER 2 is an update to that reasoning model, focused on the physical environment. Google says it builds on Gemini 3.5 Flash and improves spatial reasoning, video moment finding, video progress classification, multi-robot orchestration, and multi-step tool use.

The capability list connects perception to actions. For spatial reasoning, the model can point to objects, track them in video, detect them with bounding boxes, and plan trajectories. It can use code execution to zoom or crop images, read instruments, and annotate scenes. In task orchestration, it can combine spatial reasoning with custom robot APIs to carry out long-horizon tasks. Video features include moment finding and progress classification from continuous video feeds; the overview also names robot-video progress understanding and success detection.

The documentation presents the model as a component that can work with vision-language-action models and custom robot interfaces, rather than as a complete robot-control stack. Examples use the Google Gen AI SDK with Gemini Enterprise Agent Platform, initialize the client with `enterprise=True`, and authenticate with Google Cloud credentials, so calls are made through a user’s Google Cloud project. The separately supplied source summary says the preview endpoint is available only in the global location and uses Application Default Credentials.

The model identifier is `gemini-robotics-er-2-preview`. Google labels it Preview; the supplied material says the offering is provided as-is with limited support, while its terms allow production or commercial use. The overview directs readers to a separate model-information page for technical specifications, token and quota limits, supported capabilities, regional availability, and pricing. The supplied documentation does not give those figures directly, and it gives no publication date.

## What the archive adds
This is a new fact in a known story, not a first mention of the model family. [Our October 27 Field Note](https://newruntime.com/posts/gemini-robotics-embodied-reasoning-2-enters-public-preview-e0fdff/) covered Gemini Robotics ER 2 entering public preview; this documentation makes its capability surface and Google Cloud integration more explicit. Earlier coverage of Gemini Enterprise Agent Platform also described Google’s effort to package enterprise agent workflows behind a platform, but this page is specifically about robotics.

## Trend signal
This supports the idea that a production agent’s operational interface depends on a harness and its tools, not only on model identity. The model is presented alongside code execution and custom robot APIs for orchestration, consistent with [Harness architecture outlives model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/). The page does not establish whether these integrations are portable across models or platforms, so it is an example of the pattern rather than proof of lock-in or durability.

## Why now
The source gives no announcement date or reason for the documentation update. Its relevance now is the concrete detail it adds to the preview: a builder can see how spatial and video reasoning is intended to connect to code execution and robot APIs, while access remains marked Preview.

## Why it matters
For teams building robotics systems, the important boundary is the combination of perception, reasoning, and an interface to actual robot capabilities. Google’s examples point toward workflows in which a model interprets scenes or video, then calls tools or custom APIs as part of longer tasks. That is more actionable than treating a robotics model as a chat endpoint, but it also shifts engineering work toward testing the full loop: observations, tool calls, robot actions, and task completion.

The preview label and limited support matter for deployment planning. The page’s allowance for production or commercial use does not remove the operational uncertainty of a preview service. Teams would need to validate behavior on their hardware and environment, understand quotas and price from the linked model-information page, and design safeguards around physical actions. For builders outside robotics, the release is a signal about multimodal orchestration, not a tool change they can apply directly this week. Earlier coverage: [Gemini Robotics Embodied Reasoning 2 enters public preview on Google Cloud](https://newruntime.com/posts/gemini-robotics-embodied-reasoning-2-enters-public-preview-e0fdff/).

## Limits and uncertainty
The supplied overview does not provide benchmark results, latency, reliability, pricing, token or quota limits, or detailed regional availability. It names capabilities but does not establish how well they perform across different robots, sensors, environments, or long-horizon tasks. It also does not say whether the model can safely control hardware without an external policy or human review. Preview access and limited support imply that operational guarantees should not be assumed. The source gives no announcement date, so the documentation alone cannot establish when the capabilities became available or what changed relative to earlier preview material.

## Reader takeaway
Gemini Robotics ER 2 is a specialized preview for connecting visual and spatial reasoning to robot tools, APIs, and multi-step tasks. Its significance lies in that integration surface, not a general improvement for everyday coding agents. Robotics teams should treat the overview as a capability map and verify the separate pricing, quota, and technical details before designing a deployment around it.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Gemini](https://newruntime.com/topics/gemini.md): Explore the gemini topic hub.
- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)