Nod itself costs nothing. The AI agent from n8n is open source under the Apache 2.0 license, and there is no n8n subscription. Costs come from three places: the model, the hardware or server that runs Nod, and possibly the electricity for a device that stays on around the clock. For the model you have three options: according to Nod, you sign in with a ChatGPT Plus or Pro subscription at no extra cost. With an API key you pay per use directly to the provider, and Claude only works this way. A local model through Ollama or LM Studio costs nothing per request but needs suitable hardware. With an API model, Nod shows the cost per task, sets a $200 monthly limit when you finish setup and checks in once a task has spent $5. As of November 15, 2026.
I installed the Nod server from source on my Mac on November 15, 2026, version 0.7.6. The prices and rules in this article come from the Nod docs and from the OpenAI and Anthropic pricing pages, retrieved on November 15, 2026. Everything I have written about AI agents is collected on my AI agents page.
What Nod costs and what it doesn't
For your own server, the provider shows you the price before you create it. The Nod docs name Hetzner Cloud and DigitalOcean for a persistent server, each with Ubuntu 24.04, and point out that Hetzner bills allocated servers even when they are powered off. According to the docs, small always-free machines have less memory than the starting size. How to set up Nod on a server or a Mac mini is covered in Nod Self-Hosted: AI Agent by n8n on Mac mini.
The three model options at a glance
According to the docs, how Nod counts your usage depends on how you pay for the model:
According to the docs, the amounts in Usage are Nod's estimates from each model's API price; your provider's bill is the final word.
ChatGPT Plus or Pro: the subscription you already have
With ChatGPT Plus or Pro you sign in to Nod directly with your ChatGPT account, at no extra cost according to Nod. Plus costs $20 a month, Pro $100, $200 or $500. According to the docs, Nod's initial choice is GPT-6.1 Sol or GPT-6 Sol at high effort. The steps are in Install Nod: Set Up the AI Agent by n8n.
It isn't free in the strict sense. According to OpenAI, a connected app draws from the same allowance your plan includes for Codex and ChatGPT Work; no extra allowance is added. In ChatGPT, under Settings → Usage, you can set a weekly limit for each app as a percentage of your overall plan usage. How big the plans' allowances are is covered in Codex Pricing and Usage Limits 2026.
In Nod, Usage hides costs, forecasts and the monthly limit when a ChatGPT plan is selected. The $5 check-in still applies: Nod counts plan usage at API prices for it and says how much of the amount is plan usage. Voice costs extra even with a subscription if you want it through OpenAI: according to the docs, OpenAI speech requires API access, separate from a ChatGPT plan. Local voices, on the other hand, are free to download.
API key: pay per use
With an API key you pay for every token Nod sends to the model and gets back. According to n8n, more than 30 providers work, including OpenAI, Anthropic, Google Gemini, OpenRouter, Amazon Bedrock, Google Vertex AI and Azure OpenAI. The official list prices of the models Nod picks first for OpenAI and Anthropic, of the small models for background work, and of Claude Sonnet 5.5 for comparison:
According to the docs, background work such as memory upkeep uses a smaller model from the same provider, such as gpt-6-luna or claude-haiku-5-5. The approval check can also cause API costs. Both count toward the monthly limit.
Claude in Nod
Claude only works in Nod with an API key. The docs say: “Anthropic's authentication conditions also require an API connection for Claude in Nod.” According to its own docs, Anthropic doesn't allow third parties to route requests through Free, Pro, Max or Ultra credentials. So for Claude in Nod and other third-party tools you need an API key; a Claude subscription doesn't help you there, and the API costs come on top. How to create a key and set a spending limit at Anthropic is covered in Claude API Key: Setup, Pricing, First Call.
OpenRouter
OpenRouter gives you many models with one key or a browser sign-in. Nod shows the input and output prices next to each OpenRouter model, in US dollars per million tokens, and marks free models. According to the docs, these are the rates from the last model list update; what you actually pay can vary with routing, caching and context length. If you choose a free OpenRouter model, Nod doesn't start a paid small checker for it.
Gemini and the EU region at OpenAI
According to the docs, Gemini 3.8 Flash has a free tier with per-project quotas; different terms apply there to what Google does with your data. Gemini 4 Argon needs paid API access. For an OpenAI project with EU data residency, you choose the region “Europe (EEA + Switzerland)” in Nod. According to its pricing page, OpenAI charges a 10 percent uplift for such regional endpoints on models released on or after March 5, 2026.
Local model: no cost per request
A local model runs on your own computer or on a server with an OpenAI-compatible interface, through Ollama, LM Studio, vLLM or llama.cpp. According to the docs it needs two things:
- A model with tool calling. Nod does all its work through tools; a model without them can only chat.
- At least 64,000 tokens of context. Nod's instructions and tools take about 20,000 tokens before your first message. According to the docs, most servers start with 4,096 tokens, and some then cut Nod's instructions off without an error.
For Ollama, the docs give ollama pull gpt-oss:20b as an example and starting it with OLLAMA_CONTEXT_LENGTH=65536 ollama serve. How to set up Ollama is covered in Setting Up Local AI: Ollama to SGLang.
Usage shows these requests as “No price”, and they count neither toward the monthly limit nor toward the $5 check-in. Still, this option isn't entirely free to run. First, you pay for hardware and electricity. Second, by default your local model checks the approvals itself, and according to the docs smaller models decide less reliably. If you pick a small model from another provider or a decision service as the checker instead, that can cause API costs. What a computer for local models costs is covered in Your Own AI Server: Hardware and Cost, and the right memory size for a Mac in Local AI on a Mac: Mini as AI Server.
Cost display and monthly limit
Under Usage (⌘5 on a Mac, Ctrl+5 on Windows and Linux), Nod shows this month's spending, the limit and a forecast with an API model, plus every model call per request with model, tokens and cost. The share of background work is shown there too. Recurring schedules show their projected monthly cost. You can export the whole history as a CSV, and in Telegram or WhatsApp /usage sends a summary.
The defaults according to the docs:
- Monthly limit of $200, set when you finish setup. If you set up Nod before this change, there is no limit until you set one. You can change it from $1 up, in Usage or in chat, for example with “Set my monthly limit to $40”.
- Check-in at $5 per task since your last message. Nod stops before the next step, states the cost and asks whether to keep going.
- Preparation allowance of $5 a month for automatic preparation, separate from the monthly limit and off for new profiles.
When the monthly limit is reached, schedules, tasks and background work stop before Nod calls the model. Work that's already running can finish, and an action you already approved still runs once. Nod then only replies to your messages to change the limit. The pause lifts at the start of next month or as soon as you raise or remove the limit.
Worked examples per option
All examples are cost examples with my own assumptions, not a measurement. For the API option:
Cost per month = tasks × calls per task × (input tokens × input price + output tokens × output price) ÷ 1,000,000
Assumptions: 10 model calls per task, each with 30,000 input tokens, of which about 20,000 are Nod's instructions and tools according to the docs, and 1,000 output tokens. Per task that is 300,000 input tokens and 10,000 output tokens. The numbers assume no caching and no check or background calls.
The formula column shows the tokens per task in millions (0.30 million input, 0.01 million output) times the price per million.
Three things stand out. With GPT-6.1 Sol, the API is cheaper than ChatGPT Plus up to just under 29 tasks a month. In this example, Nod's default monthly limit of $200 would be reached after about 285 tasks with GPT-6.1 Sol and after just over 140 with Claude Opus 5.5. And the cheapest models cost only a fraction, but are weaker at the approval check: in 240 test decisions, according to n8n, GPT-6.1 Sol and Claude Opus 5.5 got every one right, while Claude Haiku 5.5 let 8 unsafe actions through. If Nod acts for you, the docs advise picking a capable checker.
Repeated input from the cache costs only a fraction of the normal input price at OpenAI and Anthropic, for GPT-6.1 Sol $0.10 instead of $2 per million tokens. The docs don't say whether or how much Nod uses it. Nod shows the real value per task under Usage. How these costs compare with your own AI server running Claude Code on a subscription is covered in Nod vs Claude Code: AI Agents Compared.
Frequently asked questions
What does Nod cost?
Nod itself costs nothing. The software is open source under Apache 2.0, and n8n doesn't charge for it. You pay for the model, either through a ChatGPT Plus or Pro subscription, per use through an API key, or nothing per request with a local model, plus possibly a server and electricity.
Can I use Nod with ChatGPT Plus?
Yes. You sign in with your ChatGPT account, at no extra cost according to Nod. According to OpenAI, the usage draws from your plan's allowance, which you also use for Codex and ChatGPT Work. Nod then shows no costs and no monthly limit.
What does Claude cost in Nod?
Claude only works in Nod via an API key, and you pay Anthropic per use. According to Anthropic, Claude Opus 5.5 costs $4 per million input tokens and $20 output, Claude Sonnet 5.5 $2 and $10. In the cost example above, that is $1.40 and $0.70 per task. You can't use a Claude subscription in Nod.
Is Nod free with Ollama?
Per request, yes: Usage shows local models as “No price”, and they don't count toward the monthly limit. But you need hardware that runs a model with tool calling and at least 64,000 tokens of context, and you pay for electricity. If you use a model from another provider for the approval check, that can cause API costs.
What does Nod cost through OpenRouter?
As much as the chosen model costs at OpenRouter. Nod shows the prices per million tokens next to each model and marks free models. According to the docs, actual costs can vary with routing, caching and context length.
How do I avoid high costs with Nod?
Set a monthly limit in Usage that suits you; the default is $200. Keep the $5 check-in on, it stops every task before the next step. And in the first few days, check in Usage what a typical task really costs.
What to do next
Start with the option you already pay for: if you have ChatGPT Plus or Pro, sign it in to Nod and test one real task. Without a subscription, create an API key with the provider, set a monthly limit of around $20 in Nod and read in Usage after a week what a task costs. What else Nod can do is covered in Nod by n8n: What the AI Agent Can Do, and how it compares with Claude Cowork, ChatGPT Dots and Grok Bot in Claude Cowork Alternative: Nod Compared. Which data goes where with each model option is covered in Nod Privacy: What the AI Agent by n8n Shares. How I build AI employees with clear costs and approvals is something I show in my community Claude Practitioners.
