n8n vs Zapier vs Make: the self-host math

The one difference that decides this

n8n, Zapier and Make do the same job. A trigger fires, data moves between APIs, and something lands in another system. The choice between them is usually decided by one thing: the billing unit. Zapier and Make charge for every unit of work your automations perform. Self-hosted n8n charges nothing per unit of work, because you pay for a server instead.

That single difference drives everything below. A metered plan costs almost nothing when you run ten workflows a day and becomes your largest software line item when a workflow starts firing every minute. A flat server cost does the opposite: it looks expensive on day one and stops moving as volume grows. Neither model is dishonest. They just bill different people well.

What each product actually charges for

Zapier bills per task. Zapier's own definition: a task is counted whenever Zapier successfully completes a unit of work. Each successful action step in a Zap consumes one task, and some actions consume more than one. Triggers and polling do not count, failed actions do not count, and the built-in tools (Formatter, Paths, Filter, Delay, Looping, Storage) do not count. So a Zap with a trigger and three action steps costs three tasks every time it runs. As of July 2026 the free tier includes 100 tasks per month, and the paid Professional plan starts near 20 US dollars per month billed yearly, at 750 tasks.

Make bills per credit, the unit it used to call an operation. Make's documentation says each action a scenario performs consumes a certain number of credits, and most actions consume one. Reading, searching, creating, updating, deleting, transforming, aggregating and iterating each cost a credit. Routers and error handlers do not. The important word there is iterating: a module that loops over 200 rows spends roughly 200 credits, not one. As of July 2026 the free tier includes 1,000 credits per month, and the Core plan starts near 12 US dollars per month for 10,000 credits.

Self-hosted n8n bills nothing per execution. The Community edition is free and carries almost the complete feature set. Workflow runs, nodes inside a run, loop iterations and retries all cost zero, because the only meter is your server's CPU, memory and disk. Paid self-hosted tiers exist and need a license key, but they sell governance features such as single sign-on and separate environments, not throughput. If you have already set up n8n on a VPS with Docker and HTTPS, you have already paid the entire recurring bill.

The install command from n8n's own docs is short enough to make the point:

docker volume create n8n_data

docker run -it --rm \
 --name n8n \
 -p 5678:5678 \
 -e GENERIC_TIMEZONE="Europe/Berlin" \
 -e TZ="Europe/Berlin" \
 -e N8N_ENFORCE_SETTINGS_FILE_PERMISSIONS=true \
 -e N8N_RUNNERS_ENABLED=true \
 -v n8n_data:/home/node/.n8n \
 docker.n8n.io/n8nio/n8n

The editor then answers on port 5678. That command is for a first look only, because it uses the default SQLite file and --rm throws the container away when you stop it. A server you rely on wants Postgres and a reverse proxy in front.

A realistic month, counted

Vague comparisons are useless, because the answer depends entirely on your volume. So count it. Here are four automations a small company really runs, with how often each fires and how many billable steps each run performs.

[
  {
    "label": "Form to CRM to Slack",
    "runs_per_month": 1200,
    "billable_steps": 3,
    "units_per_month": "3,600"
  },
  {
    "label": "Order sync, every 15 min",
    "runs_per_month": 2880,
    "billable_steps": 4,
    "units_per_month": "11,520"
  },
  {
    "label": "AI inbox triage",
    "runs_per_month": 900,
    "billable_steps": 5,
    "units_per_month": "4,500"
  },
  {
    "label": "Daily revenue report",
    "runs_per_month": 30,
    "billable_steps": 6,
    "units_per_month": "180"
  }
]

The order sync alone spends 11,520 billable units per month, because a schedule trigger every 15 minutes fires 2880 times and each run touches 4 paid steps. The daily report, which feels like the heavy workflow because it has the most steps, spends 180. Frequency beats complexity every time, which is why estimating from the shape of a workflow instead of its schedule gets people the invoice they did not expect.

Together those four automations spend 19,800 billable units a month. On Zapier that lands you on the 20,000-task tier of a paid plan. On Make it is roughly twice the Core plan's included credits, so you buy up. On self-hosted n8n it is the same monthly server bill as running one workflow, and the server will not notice.

Now change one number. Move the order sync from every 15 minutes to every 5 minutes. Zapier and Make triple that line to 34,560 units and the invoice moves to the next tier. The n8n box adds a few percent of CPU. That asymmetry, rather than the headline price of any plan, is the real decision.

Where the crossover sits

There is no universal break-even point, but the shape is predictable.

• Below roughly 1,000 units a month, the hosted free tiers win outright. Nothing beats zero, and you have no server to patch.
• Between roughly 1,000 and 10,000 units, the entry paid plans are cheap, and the hosted tools are still the better trade, because a VPS plus your attention costs more than 12 to 20 dollars a month of it.
• Above roughly 20,000 units a month, metered billing compounds and a fixed server cost usually wins, and keeps winning as volume grows.
• Any workflow that iterates over rows (a CSV import, a list of 500 orders, a paginated API) skips those bands entirely, because one run can spend thousands of units.

That last case is where self-hosting stops being a preference. A nightly job that processes 5,000 records burns 5,000 credits on a metered plan and one execution's worth of CPU on n8n.

The same logic applies to AI steps, which is where automation volume is growing fastest. Every model call inside a hosted scenario is a billable unit on top of the model provider's own token charge, so you are metered twice for the same work. Running the Claude API inside n8n workflows leaves you paying the model provider only.

The n8n license, plainly

n8n is not open source under the usual definition, and pretending otherwise is how people get surprised later. It ships under the Sustainable Use License, which states:

You may use or modify the software only for your own internal business purposes or for non-commercial or personal use.

And on passing it on:

You may distribute the software or provide it to others only if you do so free of charge for non-commercial purposes.

Read that against what you plan to do. Running n8n to automate your own company's work is internal business purposes, which is permitted, and that covers almost everyone reading this. Running n8n as the engine behind a product you sell, or reselling hosted n8n instances to customers, is not covered by the free license and needs a commercial agreement with n8n. Check the current LICENSE.md in the n8n repository before you build a business on it, because terms change and only the text in the repository counts.

There is a second, milder catch. The free Community edition holds almost every feature, and a free registration unlocks a few more, such as folders and debug in editor. Features aimed at larger teams, including single sign-on, log streaming and separate environments, sit behind a paid license key. Those are governance features. Throughput is never the thing being sold.

What the flat server cost does not cover

The honest version of the self-hosting argument includes the parts that are not free.

You own uptime. When a hosted platform has an outage, you wait and someone else is paged. When your VPS has an outage, you fix it, at whatever hour it happens. You own upgrades, because n8n releases often and a container image you never update becomes a security problem. You own backups: the n8n encryption key and the Postgres database together are the only copy of your credentials and workflows, and losing the key makes every stored credential unreadable. Point encrypted restic backups at both from the first day rather than the first incident.

You also own the parts a hosted plan quietly includes. Zapier and Make maintain thousands of app connectors and repair them when a vendor changes an API. n8n has a large node library and a generic HTTP node that covers the rest, but a broken integration becomes your afternoon. Hosted plans also include compliance paperwork, support with a response time, and someone else's on-call rotation.

Budget the real number honestly: the server, plus about an hour a month of maintenance, plus the day you spend on the initial setup. If that total still beats the metered invoice, self-hosting is the right call. At 19,800 units a month it usually does. At 500 units a month it usually does not.

Which one should you pick

Pick Zapier when the connector list is the product. It has the widest app coverage and the gentlest learning curve, and if your automations are short and infrequent, per-task billing is genuinely cheap.

Pick Make when your logic is visual and branching. Its canvas handles routers, iterators and error handlers well, and its credits cost less per unit than Zapier's tasks at comparable tiers. Watch the iterators, since they are where credit usage grows without warning.

Pick self-hosted n8n when volume is high, when data cannot leave your own infrastructure, or when workflows call models and databases you already run. It is also the only one of the three where you can read the code path a workflow takes and change it. Budget a VPS and an hour a month, and read the license first. It earns its place on any 2026 self-hosting shortlist for the reason above: the meter stops.

FAQ

Is self-hosted n8n really free?

The Community edition is free to run, with no per-execution charge and almost the complete feature set. You still pay for the server. Features aimed at larger teams, such as single sign-on, log streaming and separate environments, need a paid license key. Your workflow volume is never metered, which is the whole reason to self-host.

How many Zapier tasks does one Zap use per run?

One task per successful action step, and some actions cost more than one. The trigger is free, polling is free, failed actions are free, and built-in tools such as Formatter, Paths, Filter and Delay are free. So a Zap with a trigger and three actions costs three tasks per run, which is 3,000 tasks if it fires 1,000 times in a month.

Do Make credits and Zapier tasks count the same way?

They are close but not identical. Both charge for each action a workflow performs. Make counts each module action as a credit, including transformations and every pass of an iterator, while Zapier does not charge for its built-in formatting and filtering tools. The gap matters most on workflows that loop over lists, where Make's per-iteration counting adds up quickly.

Can I use self-hosted n8n inside a commercial product?

Not under the free license. The Sustainable Use License permits use for your own internal business purposes or for non-commercial or personal use. Automating your own company's work is fine. Embedding n8n in a product you sell, or reselling hosted instances, needs a commercial agreement with n8n. Read the current LICENSE.md in the repository before you commit to a design.

At what volume should I switch from Zapier or Make to n8n?

There is no fixed number, but the pattern holds. Below about 1,000 billable units a month the hosted free tiers win, and above about 20,000 a month a fixed server cost usually wins. Any workflow that iterates over rows breaks that rule immediately, because a single run over 5,000 records spends 5,000 units on a metered plan and nothing extra on your own server.