automation

n8n vs Zapier in 2026: The Real 12-Month Cost of a 20-Zap Stack

August 15, 2026

By , Founder, AI Automation Builder

Zapier’s task, meaning one successful action step, is not the same unit as an n8n execution, meaning one full run of a workflow no matter how many steps it contains. Both vendors publish a plan capped at “10,000” of their own unit, and those two 10,000s are not the same amount of work. The 20-Zap stack this article prices below runs 2.7 billable actions per execution on average, so it needs 10,000 Zapier tasks but only about 3,700 n8n executions, comfortably inside n8n Cloud’s Pro plan at EUR 50 a month. Over a year, on Zapier’s own published pricing, that is the difference between $1,548 and about $660, prices as published August 2026. The n8n vs Zapier decision is not really about features. It is about which of those two meters your invoice runs on.

At the very bottom, under a few hundred runs a month, Zapier is still the better product: its free tier has no n8n equivalent, and that is the honest recommendation regardless of what the rest of this article argues. We run 200+ n8n workflows in production and 89 AI agents live, and we have spent thirteen years on the maintenance end of other people’s automation, so we have a preference. We also think most articles on this comparison get Zapier’s billing wrong in a way that flatters n8n unfairly, and we would rather fix that first.

What actually gets billed on Zapier vs n8n?

Zapier bills successful action steps. n8n bills whole workflow runs. Everything else in this comparison is downstream of those two sentences.

The detail almost every comparison article botches: Zapier does not charge for every step. Its own definition is that “a task is counted whenever Zapier successfully completes a unit of work for you. Failed actions are not counted.” A specific list of steps consumes nothing at all. Quoting Zapier directly: “triggers, polling for new data, and Zapier built-in data tools (Formatter by Zapier, Paths by Zapier, Filter by Zapier, Delay by Zapier, Looping by Zapier, Sub-Zap by Zapier, Digest by Zapier, Zapier Manager, Storage by Zapier, Tables by Zapier, Forms by Zapier).”

So a Zap that triggers on a form, filters out junk, reformats a phone number, then writes to a CRM and posts to Slack is five steps and two tasks. That is a more generous meter than its reputation suggests, and if you have been counting steps, your bill estimate is too high.

n8n counts differently. In n8n’s words, 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.”

Which meter is kinder depends entirely on how many billable actions your average workflow performs. Below about three, Zapier’s is competitive. Above that, it is not close. At an average of 2.7 billable actions per run, 2,000 Zapier tasks is roughly 740 n8n executions, which sits inside the EUR 20 Starter with three times the headroom. Every crossover point in this article is stated in both units from here on, because comparing a task count on one side to an execution count on the other is not a comparison, it is a coincidence.

What does a 20-Zap stack cost on n8n vs Zapier over 12 months?

Here is a typical 20-Zap operations stack, and the arithmetic that produces 10,000 billable tasks a month.

ZapsWhat they doBillable actions per runRuns per month, eachTasks per month
10Lead and form intake: create the CRM record, post to Slack, append a sheet row32006,000
6Billing sync: create the record, update status downstream22503,000
4Scheduled reporting and digests, five actions each5501,000
203,700 runs total10,000 tasks

Triggers and filters are already excluded, so 10,000 is the number Zapier’s meter actually sees for this stack, and 3,700 is the number n8n’s meter sees for the identical work: the same 20 Zaps, two different bills.

We can price the n8n side of this stack cleanly. n8n Cloud’s Pro plan covers 10,000 executions for EUR 50 a month on annual billing, EUR 600 a year, and this stack uses only 3,700 of that ceiling. At the exchange rate of August 2026, 20 EUR is roughly $22 and 50 EUR roughly $55, so that is about $660 a year, for a stack running at 37% of the plan it is paying for.

Zapier prices just as cleanly. Its Professional plan runs $129 a month at 10,000 tasks on annual billing, billed $1,548 a year, prices as published August 2026, verified against Zapier’s live pricing slider and its embedded plan data on August 15 2026. Line the two totals up: $1,548 a year on Zapier against about $660 a year on n8n for the identical 3,700-execution workload, roughly 2.3 times the price for the exact same 20 Zaps. Run your own numbers in the cost comparator instead of this article’s stack.

Two ceilings the price table does not show. Zap history retention is not tiered by plan the way many comparisons claim: every plan, Professional through Enterprise, keeps roughly 29 to 69 days of run history, and Zapier guarantees a maximum of 60 days of run data. Only Enterprise can customize that window. So your ability to answer “what did this do in March” expires in about two months on every plan.

Every Zap on every plan is also capped at 100 steps including everything inside paths, which is a real wall for the kind of branching workflow that grows a limb every quarter.

There is also a user question hiding in the price. Professional includes one user, and the moment a second person needs to edit anything, you are pricing Team, which starts at $69 a month annual for 2,000 tasks and includes 25 seats.

n8n includes unlimited users and unlimited active workflows on every paid tier. For a five-person ops team that difference is larger than the execution meter.

Now triple the volume, which is what a working stack does in a year. n8n at 11,100 executions has just crossed out of Pro, and the jump to Business, EUR 667 a month on annual billing, about $734 at the exchange rate above, is steep. That is more than thirteen times the Pro rate for four times the execution ceiling. Zapier’s ladder at 30,000 tasks sits between two rungs we can now verify, $129 a month at 10,000 tasks and $289 a month at 50,000, so the true number lands somewhere on that curve rather than at either end.

What we can say without guessing: past this point, self-hosting has no meter at all, and your cost becomes the server, not the volume. We broke down the full picture of what those numbers look like across a real stack in what business automation actually costs.

Why can’t I compare two automation quotes directly?

Because the number on the plan label is not the number your workload consumes, and the two vendors are not even counting the same kind of thing. This article puts the number 10,000 on both platforms, but 10,000 Zapier tasks and 10,000 n8n executions are different amounts of work whenever a workflow does more than one billable thing per run, which is most of them. The fix is mechanical: before you compare any two automation quotes, convert both to the same unit using your own workflows’ actual actions-per-run average, not the vendor’s biggest round number on the page.

Skip that step and you make exactly the mistake this stack makes on n8n: paying for a 10,000-execution ceiling while only running 3,700 through it, 37% utilization, with 63% of the plan sitting unused. That headroom is already paid for whether you use it this month or not. The same trap runs the other way on Zapier if your workflows average fewer than one billable action each. Normalize to runs first. Compare prices second.

The same rule applies past this one comparison. A vendor or agency quote built around a plan label, “10,000 tasks,” “unlimited automations,” “50 seats,” is making the same move that left this stack’s n8n plan at 37% utilization: a round number standing in for the unit of work you run.

How did Zapier’s June 2026 AI pricing change the math?

On June 15 2026 an AI step stopped costing one task. It now costs between one and five tasks per run before tool calls, and each tool call bills at the same multiplier.

Zapier’s wording is unambiguous: “Your model tier sets the base cost per run (Standard 1x, Advanced 3x, Premium 5x). Tool calls add to that base rate at the same rate.” Standard makes no tool calls; Advanced, the default for new steps, runs 3x; Premium runs 5x.

Run the arithmetic on one of them. A support triage Zap with a single AI step on the default Advanced tier, calling four tools per run, costs 3 for the base plus 12 for the tool calls. Fifteen tasks for one step. Run it 400 times a month and that step alone consumes 6,000 tasks, which is 60% of the 10,000-task plan in the stack above. The practical effect: the multiplier applies to steps that used to cost one task, so a stack that comfortably fit its plan in May 2026 can be over its cap in June without adding a single new Zap.

Going over the cap does not silently drop the work. Zapier holds the runs and lets you replay them, and with pay-per-task enabled it bills the overage at 1.25 times your per-task rate on annual billing, 2.5 times on monthly, stopping hard once usage reaches three times your plan’s allotment. The overage is survivable. It is also the month the annual price you budgeted stops being the price you pay.

Zapier has effectively confirmed the ceiling itself with its safety valve: “If a step reaches 75 tasks in a single run, your Zap pauses and asks you to approve before continuing.” A guardrail at 75 tasks tells you what a single AI step is now capable of billing.

The mitigations are real. AI steps are available on Professional, Team and Enterprise only, and connecting your own AI account drops you to a flat 1x, the move if you are committed to Zapier and running agents.

On n8n, an agent workflow that calls four tools is one execution. The tool calls are not metered by the platform. You still pay OpenAI or Anthropic for tokens, exactly as you would with your own key on Zapier, but the automation platform stops taking a multiple on top. For AI-heavy work this single change moved the crossover point down by a large margin, and it is recent enough that many of the comparisons on page one were written against the old model (SERP checked August 2026).

When is Zapier still the better choice?

At low volume with short Zaps, and when nobody on the team wants to own a server. Both of those are common and neither is a compromise.

The price advantage here is narrower than Zapier’s marketing and most comparison articles suggest, because the two sides are not counting the same unit. Professional at 750 tasks is $19.99 a month on annual billing. n8n’s Starter is EUR 20, roughly $22, for 2,500 executions. That is a wash on price, and it is the only point on the two curves where it is one. Converted, it is not even: 750 tasks is about 280 runs against Starter’s 2,500 executions.

By 2,000 Zapier tasks the same conversion applies: at 2.7 billable actions per run, that is roughly 740 n8n executions, still comfortably inside the EUR 20 Starter. Zapier Professional at that volume runs $49 a month on annual billing; Team, the tier that adds multi-user access, sits at $69 a month for the same 2,000 tasks. The price argument has already turned before most teams think it has.

The genuine Zapier win at the bottom is not the price of the paid tier, it is the free one. A hundred tasks a month, no card, no server, enough to test whether an idea is worth anything before asking for budget. n8n has no equivalent hosted free tier, and that gap matters more than twenty dollars. At that size Zapier is also simply the better product: a much deeper connector library and nothing to patch.

Not billing failed actions matters more than it sounds. An API that goes down for six hours and burns 400 retries costs nothing on Zapier. Free filters and formatters mean the defensive scaffolding around a fragile integration is free to add.

Zapier also wins outright when the maintainer is a marketing or revenue ops person with no engineering support. An automation that only one absent contractor can read is a liability regardless of what it costs to run.

What do you give up by self-hosting n8n?

n8n wins clearly at volume, on long workflows, and anywhere an AI model sits in the middle of the logic, because its advantage grows with exactly the complexity that makes Zapier’s meter spin faster: three or more billable actions per run, an agent making tool calls, or regulated data that cannot leave your infrastructure. Unlimited users on every paid tier adds to the case, a detail teams underestimate until the second or third person needs edit access and the Zapier quote changes shape.

Self-hosting is how you collect on all of that, but it has a bill of its own: the features that make a platform manageable for more than one team. It removes the execution meter, not the operational work, and the Community edition is missing a specific list worth reading before you commit.

Not included in the free self-hosted build: SSO via SAML or LDAP, meaning one company login instead of a separate password for every tool; multi-main clustering, meaning more than one server so the automation survives a machine dying; and external secrets management, meaning API keys held outside the tool. Also missing: log streaming, environments, external binary storage, projects, workflow sharing, credential sharing, and git version control. Those sit on Business and Enterprise. If your compliance requirements include SSO, the free tier does not clear the bar and the paid tier is the honest comparison.

The license boundary is narrower than the internet suggests. n8n is fair-code under the Sustainable Use License, meaning the source is visible and free to self-host for your own use, including internal business use at any company size and building workflows for clients as a consultancy. What the license blocks is reselling n8n, offering it as a hosted service to your customers, or building a competing platform on it.

Then there is the part with no license text. Somebody patches it, backs it up, and notices at 2am when it stops. Zapier’s uptime is somebody else’s job. Self-hosted, it is yours. Put a floor under what that costs: a VPS, a rented virtual server, is $10 to $20 a month at published hosting rates. On top of that sits however many hours a month someone spends patching, watching, and responding to it at 2am, multiplied by what that person’s time is actually worth to you. We are not going to invent the hours, because they depend on how many workflows you run and how well you built them. The failure mode that costs the most is rarely a crash, it is a workflow that quietly stopped firing three weeks ago while everyone assumed the data was still moving. That risk exists on both platforms, and it is why automations break silently rather than loudly that decides how much a stack really costs to own.

So how should you decide?

Count the billable actions in your five busiest Zaps, ignoring triggers, filters and formatters, and multiply by monthly runs. If you are still inside Zapier’s free 100, stay there and stop reading comparison articles. Up to about 750 the two cost the same and Zapier is the nicer place to work. Between 2,000 and 10,000 n8n is already cheaper, and the question becomes how many people need access and whether AI steps are in the picture. Above 10,000 it is not close, and self-hosting is worth costing out.

Then run the same count on the stack you expect to have in twelve months, not the one you have today. Zapier publishes its entry and Team rungs plainly, so those two numbers are knowable right now instead of at month nine. Just remember the multipliers: after June 2026, an AI step with four tool calls does not land where your step count says it will.

If the numbers say move, the real risk is sequencing rather than translation, not technical difficulty. Zaps do not import, so every workflow gets rebuilt, and rebuilding is where teams discover what a Zap was quietly doing that nobody documented. The safe pattern is to run both platforms in parallel for a cycle, move the highest-volume Zaps first because they carry the savings, and keep the Zapier account alive on a small plan until the replacements have survived a full billing period. We wrote the step-by-step version in migrating off Zapier to n8n; if the stack is large enough that a bad cutover would interrupt revenue, that ordering matters more than the tooling does, which is most of what a Zapier to n8n migration turns out to be in practice.

Pricing and platform terms verified against n8n and Zapier published documentation and Zapier’s live pricing slider, including its embedded plan data, August 15 2026. Zapier’s task-tier ladder was confirmed at its entry rung (750 tasks), 2,000 tasks (Professional and Team), 10,000 tasks, 50,000 tasks, and its highest published tier. Both vendors change pricing without much notice; confirm current figures before budgeting.

Need an automation built?

Tell us what is slowing your team down. We will scope it and send a fixed-price quote.