---
type: "post"
id: "gemini-4-carbon-is-generally-available-on-google-d0a0ba"
slug: "gemini-4-carbon-is-generally-available-on-google-d0a0ba"
title: "Gemini 4 Carbon is generally available on Google Cloud and in Agent Studio"
description: "Google Cloud’s developer guide documents Gemini 4 Carbon on Gemini Enterprise Agent Platform, with configuration in Agent Studio and access through APIs."
retrieval_nugget: "Google Cloud’s developer guide documents Gemini 4 Carbon on Gemini Enterprise Agent Platform, with configuration in Agent Studio and access through APIs. Its unified 1,048,576-token input-and-output limit, long-decode support and inline office-file inputs give builders specific capabilities to test in document-heavy or long-generation workflows."
published_at: "2026-11-11"
updated_at: "2026-11-11"
record_date: "2026-11-11"
date_kind: "published_at"
topics: ["google","gemini"]
entities: ["Google Cloud","Gemini Enterprise Agent Platform","Agent Studio","Gemini 4 Carbon","Google Gen AI SDK"]
related_patterns: ["harness-architecture-outlives-model-choice"]
novelty_label: "new_fact"
what_changed: null
prior_refs: []
source_url: "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-4-carbon"
source_urls: ["https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-4-carbon"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:gemini-4-carbon-is-generally-available-on-google-d0a0ba"
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-4-carbon-is-generally-available-on-google-d0a0ba/","markdown":"https://newruntime.com/posts/gemini-4-carbon-is-generally-available-on-google-d0a0ba.md","json":"https://newruntime.com/posts/gemini-4-carbon-is-generally-available-on-google-d0a0ba.json"}
markdown_url: "https://newruntime.com/posts/gemini-4-carbon-is-generally-available-on-google-d0a0ba.md"
---

# Gemini 4 Carbon is generally available on Google Cloud and in Agent Studio

## Retrieval answer

Google Cloud’s developer guide documents Gemini 4 Carbon on Gemini Enterprise Agent Platform, with configuration in Agent Studio and access through APIs. Its unified 1,048,576-token input-and-output limit, long-decode support and inline office-file inputs give builders specific capabilities to test in document-heavy or long-generation workflows.

## Summary
Google Cloud’s developer guide documents Gemini 4 Carbon on Gemini Enterprise Agent Platform, with configuration in Agent Studio and access through APIs. Its unified 1,048,576-token input-and-output limit, long-decode support and inline office-file inputs give builders specific capabilities to test in document-heavy or long-generation workflows.

## What happened
Google describes Gemini 4 Carbon as a frontier model in the Gemini 4 family for complex, real-world work across coding, finance, legal, cybersecurity and science. The developer guide explains how to configure the model in Agent Studio and call it through the Google Gen AI SDK or REST. A Google Cloud project and authentication using Application Default Credentials or an API key are prerequisites. In Agent Studio, users select `gemini-4-carbon` and can choose a region—Global, US or EU—set the thinking level and set an output-token limit from 32,768 to 1,048,576. The documented default output limit is 1,048,576.

The model’s unified input-and-output limit is 1,048,576 tokens. Output includes thinking, function calls and the final response, so input consumes part of the overall budget. A single foreground segment supports up to 65,536 output tokens when streaming or 32,768 when not streaming; long decodes can continue across segments up to the overall limit minus input tokens. The guide describes automatic continuation in supported SDK versions, custom loops using continuation tokens, and server-managed background decodes. For Interactions API automatic continuation, it lists `google-genai` 2.34.0 or newer for Python, `@google/genai` 2.32.0 or newer for JavaScript/TypeScript, Go 1.77.0 or newer, and Java 1.80.0 or newer. GenerateContent has its own listed minimum SDK versions.

Gemini 4 Carbon supports LOW, MEDIUM and HIGH thinking levels. MEDIUM is the default for `gemini-4-carbon`; HIGH is the default for `gemini-4-carbon-cyber-permissive`. The guide says implicit caching applies automatically to shared prompt prefixes of 4,096 tokens or more, with durable caching enabled by default. Explicit context caching is not supported.

The model accepts PDF and plain text, as well as Microsoft Office files—including DOCX, DOTX, DOC, PPTX, PPT, XLSX and XLS—and OpenDocument files (ODT, ODP and ODS). These files can be sent inline or from Cloud Storage through GenerateContent or the Interactions API. Agentic video understanding is listed as generally available, but off by default. The guide identifies Gemini 4 Carbon as a Google Frontier Model subject to Section 34 of Google’s Service Specific Terms.

## What the archive adds
New Runtime’s earlier coverage said Gemini 4 Carbon had become generally available in GitHub Copilot for eligible Pro+, Max, Business and Enterprise users, with usage-based billing ([earlier coverage, 10 November 2026](https://newruntime.com/posts/gemini-4-carbon-in-github-copilot-44e8c2/)). That was a distinct access route. This Google Cloud guide adds platform-specific configuration and technical details rather than repeating the Copilot availability story.

## Trend signal
This release supports the shift toward models being accessed through a configurable production harness rather than treated as a model name alone. Agent Studio settings, API choices, continuation behavior and SDK version requirements shape how teams can operationalize the model. That is consistent with New Runtime’s pattern on [harness architecture outliving model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/), though this guide alone does not establish a broader trend.

## Why now
The guide gives builders a concrete route to evaluate Gemini 4 Carbon in a Google Cloud project: configure a region and thinking level in Agent Studio, or call it through the documented APIs. Its version requirements also make SDK compatibility a practical first check before adopting automatic continuation.

## Connections to other events
New Runtime’s earlier coverage noted Gemini 4 Carbon’s availability in GitHub Copilot for eligible paid plans, with usage-based billing ([10 November 2026](https://newruntime.com/posts/gemini-4-carbon-in-github-copilot-44e8c2/)). Google Cloud’s guide describes a separate way to configure and call the same model. Together, these routes make the model relevant to both teams using a hosted coding assistant and teams building directly on Google’s platform; they do not imply identical pricing or terms.

## Why it matters
The most actionable details are not just the headline token limit, but how it changes implementation choices. A shared input-and-output budget means teams need to account for prompts, tool calls, thinking and generated text together. Long decodes can avoid manually stitching large outputs, but the guide’s segment limits and SDK version requirements matter when designing calls, streaming responses or recovery behavior.

Office-file inputs also make it possible to test workflows against documents teams already use, rather than converting every file to plain text first. That may simplify prototypes for document review or analysis, but it does not prove reliable performance on any particular workload. Builders should benchmark representative files, check outputs, and measure latency and cost before committing. Agentic video understanding is available but off by default, so teams should not assume it is active without configuring and testing it.

## What happens next
- **Teams adopting automatic long-decode continuation will need to align their SDK versions with Google’s documented minimums** (3 months, 75%). Watch for: projects upgrading the listed Google Gen AI SDKs before enabling continuation in production.

## Limits and uncertainty
The guide documents capabilities and setup, but it does not provide independent performance results for representative customer workloads. The stated token ceiling is a maximum, not evidence that every long request will complete successfully or economically. The material here also does not establish comparative quality against other models, real-world accuracy on office files, or the latency and costs teams should expect. Google classifies Gemini 4 Carbon as a Frontier Model subject to specified service terms; teams should review those terms and validate their own data-handling and compliance requirements before deployment.

## Skill to build now
Build a small evaluation harness for one workflow that uses long context or office files. Test representative inputs, record input and output token use, compare results across thinking levels, and verify whether your SDK meets Google’s continuation requirements. Include failure and partial-output handling in the test: a large token ceiling does not remove the need to validate completion and correctness.

## Reader takeaway
Gemini 4 Carbon’s Google Cloud guide makes long decodes, a unified million-token budget and office-file inputs concrete options for builders. Treat them as capabilities to evaluate—not proof of workload performance—and check SDK compatibility before relying on automatic continuation.

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: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- topic hub: [Gemini](https://newruntime.com/topics/gemini.md): Explore the gemini 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)