Home / News

Gemini 4 Carbon lands in n8n: stronger agents, a price that doubles soon

n8n is a workflow automation platform where the AI model is a swappable part, so a small business can move its agents to a newer or cheaper model without rebuilding the process. Google's Gemini 4 Carbon is now one of those options in n8n, and it is worth testing on agents that read customer emails or forms.

According to the n8n Blog, Carbon is available in the AI Agent node and the standalone Google Gemini node. On n8n Cloud it runs on Gateway credits. On self-hosted n8n you add your own API key. Google reports 53.9% on AutomationBench and a 0.0% attack success rate on Gray Swan's prompt injection benchmark. The price is $2 per million input tokens and $10 per million output tokens until November 30, 2026, then $4 and $20.

What it means for a small business

Prompt injection is the real risk for agents that process inbound leads, support emails or supplier invoices, because anyone can hide an instruction inside a message. Better resistance helps. Still, every number above comes from Google's own announcement, and n8n itself says no model is fully resistant. Keep tool permissions narrow and route anything that sends money or data through human approval. The sensible setup is the one n8n suggests: use Carbon only on the steps that need heavy reasoning, and keep a lighter model for classifying and formatting. Plan your budget around the post-launch price, because it doubles in under three weeks.

Why it matters: Run Carbon on one real flow with your own data. Compare the cost per processed request with your current model, and decide based on the $4/$20 price, not the launch price.

Source: n8n Blog