n8n vs Zapier: Which to Pay For in 2026

The agent worked. The bill was the surprise.

I had moved a support-triage flow from a plain classifier to an AI agent that could look things up before deciding. On one platform the monthly cost stayed exactly where it was. On the other it started tracking how curious the agent felt that week, because every lookup was a billable unit. Same logic, same inbox. That gap is the whole reason this comparison exists.

The n8n vs Zapier articles ranking for this keyword run from roughly 2,800 to 5,500 words, and they agree on the basics: Zapier is easier and connects more apps, n8n is cheaper at volume and can be self-hosted, and Zapier bills per task while n8n bills per execution. All of that is true. None of them walks through what happens to either bill once the workflow contains an agent that decides for itself how many steps to take, and only one mentions that n8n's self-hosted license has commercial limits, without saying which ones bite. Both are below.

For the wider field (Make, Lindy, Relay.app and others), see our AI workflow automation roundup. If you are still deciding whether you want an agent at all, start with our guide to AI agents.

What each one actually is

Zapier is a hosted automation service. You pick a trigger, add actions, and it runs in Zapier's cloud. Its app directory says it connects 10,234+ apps, and for a non-technical team that number is the product: whatever obscure CRM your sales lead bought, there is probably a Zapier integration for it. Zapier has added AI steps, a separate Agents product, and an MCP server that lets outside AI tools run Zapier actions.

n8n is a node-based workflow builder you can run on n8n's cloud or on your own server. Its integrations page lists 2,329 integrations, one of which is the HTTP Request node, so in practice anything with an API is reachable if you are willing to read the API docs. You can drop JavaScript or Python into a Code node. The AI Agent node connects to whichever model provider you give it credentials for, including local models.

Zapier is for people who want the automation. n8n is for people who want to own it.

The short answer

	Zapier	n8n
Hosting	Zapier's cloud only	n8n Cloud, or self-hosted
Integrations listed	10,234+ apps	2,329, plus HTTP Request for any API
Billing unit	Task: each successful action step	Execution: one run of the whole workflow
AI agent billing	Agents use "activities," a separate pool	Agent runs inside the workflow's one execution
Free option	100 tasks a month	Self-hosted Community Edition
Entry paid plan	Pro: $19.99 a month billed yearly, $29.99 monthly, 750 tasks	Starter: €20 a month billed yearly, €24 monthly, 2,500 executions
Next tier	Team: $69 a month billed yearly, 2,000 tasks	Pro: €50 a month billed yearly, 10,000 executions
Over the limit	Overage at 1.25x (annual) or 2.5x (monthly), pausing at 3x your allowance	Upgrade the plan

Prices are from zapier.com and n8n.io as of October 2026.

How a task differs from an execution

On Zapier, a task is counted every time a step successfully does work. A standard action is a task. So is an AI by Zapier step, a Code by Zapier step, and a call made through Zapier MCP. Triggers are free, and so are Zapier's built-in utilities: Filter, Paths, Formatter, Delay, Looping and a few others. Failed actions are not counted.

On n8n, an execution is one production run of a workflow, and n8n's pricing page is blunt about it: it does not matter how many steps it has or how much data it processes. Manual test runs from the editor do not count. Neither do sub-workflows called with the Execute Sub-workflow node; only the parent run counts.

That means a 12-node n8n workflow and a 3-node one cost the same per run. On Zapier, a Zap with eleven actions costs eleven times what a one-action Zap does.

Price it with one real workflow

Take a support inbox getting 3,000 emails a month. Each email triggers a flow that classifies it with an LLM, creates a ticket, and posts urgent ones to Slack. Call it three actions.

On Zapier that is up to 9,000 tasks a month (three per email, fewer when the Slack step is filtered out, since filtered steps never run). That is twelve times Pro's 750-task entry allowance, so you are buying a much higher task tier, and if you guess low on a monthly plan, the overage runs at 2.5x the base rate.

On n8n it is 3,000 executions. Which, annoyingly, is 500 more than Starter includes. You land on Pro at €50 a month billed yearly, and the count would not move if the flow grew to ten steps.

What an AI agent does to each bill

Now replace the classifier with an agent. Before it labels an email, it can look up the sender in the CRM, search the help docs, and check whether the customer has an open ticket. Some emails need none of that. Some need all three lookups and a second search.

On Zapier there are two ways to build this, and both meter the agent's behaviour. As a Zapier Agent, usage is counted in activities, a pool separate from tasks: using a trigger, answering from a knowledge source, running an action, a web search and web browsing each count. Pro includes 1,500 activities a month. If you build it inside a Zap instead, with Looping by Zapier over the lookups, the loop itself is free but every action that runs inside it is a successful step, and each one is a task.

On n8n, the AI Agent node calls its tools inside the run that triggered it. One email, one execution, whether the agent made zero tool calls or seven. If the tools are themselves sub-workflows, those still do not count.

Here is the catch on the n8n side: the model tokens are yours. n8n's agent runs on whatever provider credentials you connect, so every extra tool call is another round trip billed by OpenAI, Anthropic or whoever you chose. The meter moves. It just moves on a different invoice, priced by the token instead of by the step, and if you point the agent at a local model through Ollama (our Ollama vs LM Studio comparison covers that setup), the marginal cost of a curious agent gets close to zero.

My rule now: if you cannot predict how many steps the agent will take, do not put it on a plan that bills by the step.

The n8n license clause freelancers miss

n8n's self-hosted Community Edition is free under the Sustainable Use License, which n8n calls fair-code. The OSI would not class it as open source. The operative sentence: you may use or modify the software "only for your own internal business purposes or for non-commercial or personal use."

For a company automating its own operations, that is no restriction at all. For a freelancer or a small agency it matters, and n8n's license FAQ draws the line in a specific place. You can charge clients for building, setting up and maintaining workflows, and n8n's own example is a consultant running workflows for several small-business clients on a single instance, where clients pay for outcomes rather than n8n access. You cannot host n8n as a service and let clients build or edit their own workflows. Embedding a workflow builder in your SaaS product so your users can create automations also needs an Enterprise license.

So the test is who touches the editor. If only you do, you are fine. If your client logs in to tweak a node, you have crossed the line, and the FAQ points you to license@n8n.io.

Zapier has no equivalent question. You pay per account, and that is the end of it.

Where each one loses

Zapier loses on cost once workflows get long or agentic, and its linear editor gets awkward with heavy branching. It is also cloud-only, which settles the question for any team whose data cannot leave its own servers.

n8n loses on time. Self-hosting means you own the updates and the 2 a.m. outage, and SSO on n8n's paid tiers starts at the €667-a-month Business plan, which is self-hosted only. The editor is also less forgiving. A marketing coordinator can ship a working Zap before lunch; the same person in n8n will probably be reading about JSON expressions by mid-afternoon.

Which one to pay for

Pay for Zapier if the people building automations are not developers and the app you need is one of the thousands n8n has no node for. Pro is enough to find out. Switch to annual billing before you hit overage, since the annual rate is half the monthly multiplier.

Pay for n8n if someone on the team reads API docs without flinching and your workflows run past five or six steps, and especially if you are building agents whose tool use you cannot cap in advance. Start on Cloud Starter, move to Pro the month you pass 2,500 runs, and self-host only once somebody has agreed to own the server.

I run client-facing agents on n8n and leave the dull glue (form to spreadsheet, spreadsheet to Slack) on Zapier, where nobody has to think about it.

Frequently Asked Questions

Is n8n cheaper than Zapier?

Usually, once workflows have more than a couple of steps. Zapier counts every successful action as a task, while n8n counts a whole workflow run as one execution no matter how many steps it has. Zapier Pro starts at $19.99 a month billed yearly for 750 tasks; n8n Cloud Starter is €20 a month billed yearly for 2,500 executions.

How does Zapier bill AI agents?

Zapier Agents use activities, a pool separate from tasks. Using a trigger, answering from a knowledge source, running an action, a web search and web browsing each count as an activity. Pro includes 1,500 activities a month. AI by Zapier steps inside a regular Zap count as tasks.

Do n8n AI agent tool calls count as extra executions?

No. The AI Agent node runs its tool calls inside the execution that triggered the workflow, and sub-workflows called with Execute Sub-workflow do not count toward the limit. You still pay your model provider for the tokens each tool call uses.

Can I self-host n8n for clients?

You can build and maintain workflows for clients and charge for that work, including on a single instance you run. You cannot let clients build or edit their own workflows on an n8n you host, or embed n8n's builder in a product for your users, without an Enterprise license from n8n.