Both of these automate work across the apps you already use, and both price you by the unit of work. The unit is where the comparison begins, because the two vendors count different things and neither publishes a conversion between them. n8n counts the run. Zapier counts each successful action inside the run. The buyer question on this page is what the same automation costs on each meter, with the scenarios where the meters part ways made explicit.
The meter is the axis
n8n's definition is one sentence on its pricing page: an execution is a single run of your entire workflow, and it does not matter how many steps are in the workflow or how much data it processes, it is still a single execution. The entry cards repeat the shape as workflow executions with unlimited steps. Zapier's Help Center defines a task as any successful action that runs in Zapier, and says only successful actions count toward task usage. One is a whole-run unit, the other a per-action unit, and that difference is why the two allowances in the table are not two counts of the same thing.
One run means different numbers on each platform
Take the same trigger firing once, in the simplest shape where it leads to one billable action. Both vendors' definitions cover it: n8n counts one execution and Zapier counts one task. The accounts agree, and this is the case where the meters look identical.
Take the trigger firing once and performing three successful actions. n8n still counts one execution, because step count is explicitly outside its meter. Zapier counts three tasks, because each successful action is a task. The same automation, two different bills, and the difference comes straight from the two definitions in the table.
What happens when the workflow branches
A branch that does not run is billed by neither vendor in the way the definitions read. Zapier's own article lists steps that do not run as not counting, and its Filter or Paths steps as not counting at all, so the path that does not execute adds no tasks. n8n counts the run regardless, and branching inside one run does not change the execution figure. The asymmetry is the shape, not a fault: on Zapier the bill follows the actions that actually succeed, on n8n the bill follows the run.
What happens when it loops
A loop is where the temptation is to write a false universal, and this page will not. Zapier does not bill one task per loop item by definition. It bills each successful action performed per item. A loop over N items doing one successful billable action each consumes N tasks; the same loop doing two successful actions per item consumes 2N tasks. That is the definition restated, not a new formula. On n8n, the whole loop is one execution only while it stays inside one workflow run, which the vendor's own meter definition supports. Neither vendor publishes a fixed ratio between loop items and price in either direction, so any worked example here is illustrative and says so.
Where AI changes the meter
The two platforms do not line up cleanly here, and the honest move is to keep the products separate rather than force a conversion. n8n has several surfaces: normal workflow executions, Assistant credits that pay for n8n Assistant and the AI Workflow Builder on Cloud, and an AI Agent node that runs inside workflows. Zapier has its normal task meter, AI by Zapier inside the task economy, and the legacy Agents product with its own activity meter. Only Zapier publishes a complete AI pricing formula, and only n8n publishes what a credit is worth without a consumption rate. So there is no apples-to-apples AI cost comparison on this page, and trying to build one from the published pages would be inventing the missing half.
What failure and replay do to usage
Only what both vendors document is quantified here. n8n's pricing page says manual test runs, sub-workflows and error workflows do not count toward the execution meter. Zapier's article says only successful actions count, action steps that error or halt do not, and previously successful steps that run again when you replay a Zap count as one task per successful run. What happens to a retried run on each platform beyond those published sentences is not stated by the vendors in comparable language, so this page does not put a number on it.
Hosted versus self-hosted
Zapier is a hosted platform and documents no self-hosting or on-premises option. n8n splits the question into three hosting shapes. The self-hosted Community edition is free under the Sustainable Use License. Business is currently self-hosted only. Enterprise can be hosted by n8n or self-hosted, and its card reads Contact Sales. Paid self-hosted editions use license keys. That means the n8n choice is also an infrastructure choice. The person who picks self-hosted n8n owns the servers, the updates, the security patching, the backups, the uptime and the monitoring, and on a paid self-hosted edition the license key must ping the vendor daily to stay active. With Zapier, the vendor operates the hosted platform, so you do not run or maintain the automation infrastructure yourself. Self-hosting is a real option, not automatically a better one.
What you can actually take with you
Exporting a JSON file is not the same as being able to run it somewhere else, and this is where the two part most visibly. n8n documents workflow export and import as JSON, the docs say exported files include credential names and IDs, and the CLI has separate commands for workflows and credentials. The exported workflow can be re-hosted on your own n8n instance, which is the only thing that makes portable a literal word. Zapier's JSON import and export is a backup and share format: the current help article limits it to Team and Enterprise accounts, says import is not available during a free trial, and export is further restricted by ownership for non-admins. Nothing that leaves Zapier can run outside Zapier, because there is no self-hosted Zapier.
Pricing without pretending the units are equal
The table shows each vendor's own published figures, and comparing them requires keeping the meters separate. Zapier's $29.99 Professional starts at 750 tasks a month, and its own Help Center counts a task as one successful action. n8n's 24€ Starter includes 2.5k executions with unlimited steps, and its own FAQ counts an execution as one run regardless of complexity. Neither figure converts into the other, and no universal per-execution versus per-task ratio is published by either vendor. Where a specific automation makes the arithmetic worth doing, the workflow shape has to be stated first, which is why the scenario sections above name the shape before the count.
Scorecard
The five components and their weights are the same ones our methodology publishes for every review. Each dimension below names what it is judging, the facts it rests on, and where the score is an editorial call rather than a published number.
Fit for the job, 30 percent. The judgment here is which meter answers the buying question a buyer is most likely to be asking, and the two answers serve different buyers. n8n's meter makes a complex run predictable: step count, branching, loops and data volume inside one run do not change the figure, and the unlimited-steps wording makes the entry plan legible. Zapier's meter makes every action estimable: you can count the actions a Zap does and know the bill, and its published list of non-counting steps sharpens the estimate. On the axis of a buyer who wants one number for a whole workflow, n8n scores higher. On the axis of a buyer who wants per-action transparency, Zapier does. No single winner is forced; the fit scores are 24 of 30 for n8n and 21 of 30 for Zapier because the editing call here weighs the whole-run meter for a branched workflow. That is a judgment about which meter is more predictable for a workflow with multiple actions in one run, not a claim that either platform automates more.
Cost to get started, 25 percent. n8n's 24€ entry includes 2.5k executions with unlimited steps, and the self-hosted Community edition is not sold under the paid monthly execution allowance shown on the subscription cards. The trade-off: whether that cost lands depends on a team's ability to self-host, and the paid Business tier is self-hosted only. Zapier's $29.99 entry includes 750 tasks, its free tier starts at 100 tasks, and every successful billable action counts as a task. Neither is cheaper as a claim, because the units do not convert. Scored on published entry price and included work within each vendor's own unit: n8n 20 of 25, Zapier 18 of 25. The n8n figure rests mainly on the standing free self-hosted edition under the Sustainable Use License, a $0 option that exists all year rather than a trial; Zapier's estimate is only slightly lower because its free tier is smaller within its own unit and every successful billable action counts as a task.
What the vendor commits to, 20 percent. Both vendors publish a written accounting model: n8n a pricing FAQ, a changelog and a license text; Zapier a Help Center that defines the task, the non-counting steps, replay, pay-per-task and the AI formula. Zapier's set is more complete on the AI side because it publishes the formula; n8n publishes a credit value without a consumption rate. n8n 15 of 20, Zapier 17 of 20.
Portability, 15 percent. n8n's JSON export plus a self-hosting route makes portability literal, with documented costs (credentials move separately, paid self-hosted editions ping the vendor daily). Zapier's JSON export is gated to Team and Enterprise, ownership-restricted, and cannot run outside Zapier. n8n 14 of 15, Zapier 9 of 15.
Documentation quality, 10 percent. n8n's pricing FAQ, docs and Help Center answer quota, credits, license and region in writing; the n8n review notes several footer and legal URLs no longer resolve. Zapier's Help Center covers task accounting, replay, pay-per-task and AI formula with dated updates. n8n 8 of 10, Zapier 9 of 10.
Weighted at the published weights, n8n scores 81 and Zapier scores 74. The totals matter less than the fit scores, because neither total overrides the buyer shape each product is built for.
For the buyer who wants one price per run
You are technical enough to run software yourself, and you want one price for the run no matter how many steps it has. The 24€ Starter includes 2.5k executions with unlimited steps, the free self-hosted Community edition puts a real option on the table, and the whole-run meter makes branching and loop-heavy automation predictable. The cost is that you take on the infrastructure work, and that the paid Business tier is currently self-hosted only.
For the buyer who wants per-action pricing
You want to start in the browser without running anything, and you need every action priced on its own. A simple Zap has a simple bill, the published non-counting list sharpens the estimate, and pay-per-task takes over after your plan limit with a published ceiling. Its JSON import and export only unlock on Team and Enterprise, so if portability between platforms matters, that boundary belongs in the decision.
Neither is universally cheaper
The entry prices do not convert, and the two meters reward different workflow shapes. n8n prices the run, so branching and loops inside one run do not change what a single execution costs on the card. Zapier prices the successful action, so a Zap that fires rarely with few actions each is straightforward to estimate in tasks. Which meter fits your automation is a function of the workflow shape, not of one vendor being cheaper. The profiles carry the same figures with the same dates: n8n and Zapier, the n8n review reads the cloud-versus-self-host decision, and AI agents is the category hub this comparison belongs to.