n8n Pricing 2026: Cloud Plans and Self-Hosted Costs

n8n bills per execution, not per step. One execution is a single run of an entire workflow, however many steps it contains — a 40-step automation costs the same as a 2-step one. n8n Cloud is €20/month (Starter, 2,500 executions), €50/month (Pro, 10,000) and €667/month (Business, 40,000, with SSO), all billed annually. The self-hosted Community Edition is free and open source; you pay for infrastructure and the engineering time to run it.

Every figure on this page was rechecked against n8n.io/pricing on 7 September 2026. The Cloud tiers have not moved since this guide was first published in July 2026 — same three prices, same execution limits. n8n lists in euros and has restructured its plans before, so confirm the live page before you commit budget.

That execution-based model is the single most important thing to understand, because it inverts how most teams estimate cost. On Zapier or Make, complexity is expensive — every step burns quota. On n8n, complexity is nearly free and frequency is what you pay for. Below is how the plans actually break down, where the hidden costs sit (especially when you self-host), what an AI agent really costs to run on it, and how to decide which route fits an operations-heavy business.

How does n8n count an execution?

Per n8n's own definition, an execution is "a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes — it's still a single execution." This is the detail that trips people up — the top organic result for "n8n pricing" sits right next to a Reddit thread literally titled "N8N Pricing Confusion."

The practical consequences:

• Deep workflows are cheap. A single automation that pulls a record, enriches it across five APIs, branches, formats, and writes to three destinations is one execution. On per-task tools that same logic could be 10+ billable operations.
• High-frequency triggers are what add up. A workflow that fires on every inbound webhook, every new row, or every minute on a schedule is where your execution count climbs. Consolidate chatty triggers and batch where you can.
• Sub-workflows count too. If you call a sub-workflow, that's typically its own execution. Reusable "helper" workflows are great for maintainability but do register against your quota.

So the estimating question isn't "how complex is my automation?" — it's "how many times a month does it run?" Count your triggers and their frequency, not your nodes.

What do the n8n Cloud plans cost?

n8n Cloud is the managed, hosted option — no servers to run. The four published tiers (annual billing, in euros) look like this:

Three things worth stating plainly, because most pricing posts skip them:

• Every price above assumes annual billing. n8n's pricing toggle labels the annual option a 17% saving, which puts month-to-month at roughly €24 for Starter and €60 for Pro on that arithmetic. The monthly figures are not printed on the page — you have to flip the toggle to see them, which is worth doing before you budget a three-month pilot.
• Currency and repricing. n8n publishes in euros and has restructured this pricing more than once — third-party guides quoting the Business tier at "$800/month" aren't wrong, they're just a different currency and a different snapshot. Treat every number here as directional and confirm on the live pricing page before you commit budget.
• There is a startup discount. n8n's pricing page offers companies with fewer than 20 employees 50% off the Business plan. If SSO is what pushes you up to Business, check whether you qualify before you budget the full €667.

The jump from Pro (€50) to Business (€667) is steep and catches teams off guard. In practice you don't move to Business for the extra executions, and not for throughput either — concurrency goes from 20 to 30, which is not a 13× difference in anything. You move for SSO/SAML, LDAP and Git-based version control. If your blocker is "we need single sign-on and audit-grade access control," that's the line item you're actually buying.

Is self-hosted n8n really free?

The software is. The Community Edition is open source and free to run — that's genuinely the headline, and it's why n8n is popular with technical teams. But "free software" is not "free to operate," and this is where budgets go wrong. Self-hosting swaps a predictable subscription for an operating cost that you now own.

At ATI we frame the true cost of running any automation on our AI agent cost page as a three-layer system, and n8n self-hosting maps onto it cleanly:

The runtime layer (hosting) is the one everyone budgets for and the one that's genuinely cheap — a modest server runs a self-hosted instance fine at low volume. The operations and people layers are the ones that get underestimated. Someone has to own upgrades, watch for failed executions at 2 a.m., patch the box, and rotate credentials. When a critical workflow silently stops firing, the cost of that outage dwarfs any subscription you avoided.

That's the honest trade: self-hosting is cheapest in dollars and most expensive in attention. It wins when you have engineering capacity and a reason to keep data in your own environment. It loses when the team running it doesn't actually have room to operate it.

Self-hosted Community vs self-hosted Enterprise — what's the difference?

Confusingly, "self-hosted" isn't a single thing. There are two self-hosted editions:

• Community Edition — free, open source, the standard self-hosted n8n. Everything you need to build and run workflows.
• Self-Hosted Enterprise — a paid license that adds the governance layer on top of your own infrastructure: SSO/SAML/LDAP, external secret stores, log streaming, extended data retention, and dedicated SLA-backed support.

So the decision isn't binary "cloud vs self-host." It's a grid: managed vs self-run on one axis, community vs enterprise features on the other. Most operations-heavy SMBs start on Community self-hosted or Cloud Starter/Pro, and only reach for an Enterprise license when compliance or identity requirements force it.

Cloud or self-hosted — which should your business choose?

The deciding factors, in order of how often they actually settle it:

• Data residency / compliance. If regulated data can't leave your environment, self-hosting is the answer regardless of cost. This decides it more often than price does.
• Engineering capacity. No spare hands to run infrastructure? Cloud, full stop. The subscription is cheaper than a distracted engineer.
• Volume. At very high execution counts, self-hosting's flat infrastructure cost eventually beats per-tier cloud pricing — but only once you're paying for the ops discipline anyway.
• Governance needs. SSO, audit trails, and access control push you to Cloud Business or a self-hosted Enterprise license either way.

When does self-hosting stop being cheaper?

Not at an execution count. The subscription you avoid is a fixed number. The operating cost you take on scales with how many workflows are load-bearing and how much governance you need around them. An instance running a high volume of one internal report stays cheap. A low-volume instance that issues your invoices does not, because that one has to be watched.

Four things move the line, roughly in the order they arrive:

1. A workflow becomes load-bearing. The moment a failed execution means a customer isn't invoiced, a lead isn't routed, or a shift isn't staffed, you have taken on an on-call obligation. Failure alerting and someone who answers the alert stop being optional — that is the operations layer in the diagram above.
2. The instance stops being one box. Scaling past a single container brings queue workers, a database you now back up, and execution history you now retain. Each is another thing to run, patch, and restore.
3. You need the governance features. This is the clearest crossover, because it is arithmetic rather than judgement. SSO/SAML, an external secret store and log streaming aren't in the Community Edition, so reaching them means a paid self-hosted Enterprise licence. You then pay n8n a licence fee and carry the operations layer. Unless data residency is the reason you self-host, Cloud Business does that job on one bill.
4. Someone has to review the spend. If your workflows call models, the token bill drifts on its own. Our guidance on what AI agents cost to run is that weekly is the right review cadence for an active deployment, because monthly is too slow to catch token leakage or tool-call drift before it compounds. That review is recurring work no matter who hosts the runtime.

What keeps self-hosting genuinely cheaper is narrower than most teams assume: internal workflows where a missed run is an inconvenience rather than an incident, on infrastructure someone already patches for other reasons. That describes a lot of real automation, and those teams should self-host. It stops describing your setup the first time an operations process depends on the instance staying up.

There is a third option teams rarely price properly: keep the free software, and pay someone else for the operations and people layers. That is what a managed retainer buys. Comparing what AI automation agencies charge against the true cost of a self-hosted instance is a fairer test than comparing it against a €50 Cloud plan.

How does n8n pricing compare to Zapier and Make?

This is where the execution model pays off. Named honestly:

• Zapier charges per task — each step that runs is billable. Multi-step Zaps get expensive fast, and pricing scales with workflow complexity.
• Make charges per operation — similar dynamic, priced per module run, cheaper per unit than Zapier but still complexity-sensitive.
• n8n charges per execution — the whole workflow, once. Complexity is effectively free; you pay for frequency.

The takeaway isn't "n8n is always cheaper." It's that n8n gets relatively cheaper the more complex your automations are, and per-task tools stay competitive for simple, low-frequency zaps. If your workflows are deep — enrichment chains, multi-branch logic, heavy data transformation — n8n's model is structurally in your favor. If they're shallow and infrequent, the difference is smaller than the switching cost.

What does it cost to run AI agents on n8n?

The plan table doesn't answer this, because an agent produces three separate bills and only one of them comes from n8n.

1. The workflow runs. An AI agent workflow bills like any other workflow: one execution per run, whatever happens inside it. A chain of model calls, tool lookups and retries is still a single execution against your quota. On the n8n side, agents are not a premium product.

2. The model tokens. The model is not included. n8n's AI nodes run on credentials you supply — the documented prerequisite for the OpenAI node is your own OpenAI account and API key — so token spend lands on your model provider's invoice, not n8n's. This is the variable line, and it tracks prompt size and call volume rather than execution count.

3. AI Assistant credits. The AI credits attached to the Cloud plans pay for neither of the above. They meter n8n's AI Assistant, the helper that drafts and debugs workflows for you. n8n's pricing page states that credits are consumed as you use the Assistant, that the balance refreshes every month, and that unused credits don't roll over; Starter includes 2,300 a month and Pro up to 13,700, with the Assistant still marked early preview. Self-hosted instances have no credit allowance — n8n's deployment docs say the Assistant stays switched off until you add your own AI provider key, so that cost also goes to your provider.

For budgeting, the split that matters is fixed versus variable. The n8n bill for an agent is fixed and predictable. The token bill behind it is neither, and once the agent runs daily it is usually the larger of the two. A team that prices an agent off the n8n plan alone has budgeted the smaller number. Our breakdown of what an AI agent costs works through the same split in more detail.

How should an operations team actually budget for n8n?

A practitioner's estimating sequence:

1. Count trigger frequency, not steps. List each workflow and how many times a month it fires. Sum those — that's your execution estimate. Add headroom for sub-workflow calls.
2. Map to a plan. Under ~2,500/mo, Starter. Under 10,000, Pro. Above that, or if you need SSO, Business or self-host.
3. Price the layer you're avoiding. If you lean self-host to save the subscription, honestly cost the runtime, operations, and people layers above. Cheap software, real operating cost.
4. Budget model spend on its own line. If your workflows call an LLM, the n8n plan covers the runs and your model provider bills the tokens. Estimate that separately — it is the line that grows with usage, and it is invisible in every n8n plan comparison.
5. Tie it to outcome, not tool spend. The number that matters isn't the n8n bill — it's hours recovered. Our AI ROI calculator models that directly: team size × weekly hours lost × 4.33 × cost per hour. If a €50/month Pro plan automates away 40 hours of manual work a month, the subscription is a rounding error.

That last point is the one implementation partners see constantly: teams optimize the €50 tool bill and ignore the €5,000 of manual labor the tool was supposed to eliminate. Budget the workflow's value first, then the license.

Where an implementation partner fits

Plenty of teams run n8n themselves and do it well — the platform is built for that. Where ATI tends to get pulled in is when the automation moves from "handy internal script" to production operations: when a failed execution means a customer doesn't get invoiced, when execution counts spike unexpectedly, or when the self-hosted instance no one has time to babysit becomes a single point of failure. We design, ship, and operate automation systems with the cost architecture, monitoring, and governance that keep them reliable at scale — the operations and people layers, handled. If you're weighing whether n8n is the right foundation or whether the ops burden makes managed delivery smarter, book a strategy call and we'll map it against your actual workflows.

Frequently asked questions

How much does n8n cost per month?

n8n Cloud is €20/month for Starter (2,500 executions), €50/month for Pro (10,000 executions) and €667/month for Business (40,000 executions, with SSO), all billed annually; Enterprise is custom-priced. Those three figures were verified against n8n.io/pricing on 7 September 2026 and are unchanged since July 2026. The self-hosted Community Edition is free — you pay only for infrastructure and operations. n8n publishes in euros and has repriced before, so confirm the live page before budgeting.

What counts as an execution in n8n?

One execution is a single run of an entire workflow, regardless of how many steps it contains or how much data it processes. A 40-step workflow and a 2-step workflow each count as one execution per run. This is different from Zapier (per task) and Make (per operation), which bill for each step.

Is self-hosted n8n free?

The Community Edition software is free and open source. Operating it is not: you pay for hosting/infrastructure, plus the engineering time to maintain, monitor, secure, and upgrade it. For low-volume use with available engineering capacity it's very cost-effective; without that capacity, Cloud is usually cheaper in total.

When does self-hosting n8n stop being cheaper?

Not at a particular execution count. It stops being cheaper when a workflow becomes load-bearing enough to need failure alerting and someone on call, when the instance grows past a single container into queue workers and a database you back up, or when you need SSO, an external secret store or log streaming — those sit behind a paid self-hosted Enterprise licence, so you end up paying n8n and carrying the operating cost. Self-hosting stays cheapest for internal workflows where a missed run is an inconvenience, on infrastructure someone already maintains.

Is n8n cheaper than Zapier?

It depends on your workflows. Because n8n charges per full execution rather than per step, it gets relatively cheaper as automations grow more complex. For deep, multi-step workflows n8n's model is structurally cheaper; for simple, infrequent automations the gap narrows and switching cost may outweigh savings.

Which n8n plan do I need for SSO?

SSO/SAML/LDAP is available on the Cloud Business plan and on the self-hosted Enterprise license. If single sign-on and audit-grade access control are your requirement, that's the tier you're buying — not the extra execution capacity.

Do AI agents cost extra on n8n?

Not on the n8n bill. An AI agent workflow counts as one execution per run, the same as any other workflow. What costs extra is the model: n8n's AI nodes use API credentials you provide, so token spend goes to your model provider. The AI credits included on Cloud plans (2,300/month on Starter, up to 13,700 on Pro) are separate again — they meter n8n's AI Assistant for building workflows, not the models your workflows call.

How do I estimate my n8n execution usage?

Count how often each workflow runs per month based on its trigger frequency — not the number of steps. Sum those runs, add headroom for any sub-workflow calls (which count separately), and match the total to a plan tier. Frequency drives cost far more than complexity does.