Google Gemini 4 Carbon Comes to n8n With a Focus on Safer Agents

A model aimed at complex automation

Gemini 4 Carbon is available in n8n through the AI Agent node and the standalone Google Gemini node. n8n reports that Google’s model achieved 53.9% on AutomationBench, a benchmark for automation work, placing it at the top of the results reported by Google.

The model is designed for workflows that require multiple steps and tool calls. According to Google, internal team runs have lasted for days while working through complex engineering problems. n8n says this could allow agents to continue through more stages before losing the thread.

Security and prompt injection resistance

Security is a central focus of the launch. According to results reported by Google, Gemini 4 Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for checking whether vulnerabilities are exploitable, and 71.0% on CWE-bench for patching them.

n8n also highlights the model’s result on Gray Swan’s prompt injection benchmark. Carbon recorded a 0.0% success rate for attacks, compared with 0.7% for Gemini 4 Argon. This matters for agents that process untrusted content, including inbound emails, web-hook payloads, form submissions and scraped pages. n8n cautions that no model is completely resistant to malicious attacks at scale, so users should keep their security protocols in place.

One example workflow starts with a security alert received through a web-hook. An agent running on Gemini 4 Carbon can read the alert, retrieve related logs, check known issues, assign a priority and post a summary to Slack for an on-call engineer.

How n8n users can access Carbon

To use the model, users can open an AI Agent node, add the Google Gemini Chat Model and select Gemini 4 Carbon from the model list. The model is also available in the standalone Google Gemini node, where users can connect approved tools and add memory when an agent needs additional context.

On n8n Cloud, Carbon can be used through Gateway credits without a separate provider account or extra billing. Self-hosted users need to add their own Gemini API key.

n8n says users already working with Gemini 4 Argon do not need to move every workflow to Carbon. Carbon can handle steps requiring heavier reasoning, while Argon or a lighter model can remain in use for simpler tasks such as classification or formatting.

Pricing and what it means for workflows

Google launched Carbon at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens. According to n8n, those prices will rise to $4 and $20 after the introductory period ends on November 30, 2026.

For n8n users, the launch provides another option for agents that must work through lengthy tasks or handle potentially hostile input. The model’s reported benchmark results support its use in security and automation workflows, while n8n’s guidance suggests combining Carbon with lighter models rather than using it for every task.