n8n vs Zapier Pricing: What 10,000 Runs Actually Cost

n8n vs Zapier pricing comparisons fail in the first line, because they put two numbers side by side that are measured in different units. Zapier sells tasks. n8n sells executions. Those words look like synonyms and they are not: one is a step, the other is an entire run. Comparing “$29 for 750” against “$24 for 2,500” without converting the units first is like comparing a gas price in dollars-per-gallon to one in dollars-per-mile and declaring a winner.

I run n8n in production. The workflows behind my social pipeline, my email capture and my daily video all live there, and I have watched the execution counter long enough to know exactly where it spikes. The number that made this concrete was a content batch job: one nightly workflow that loops over a list of items and does three things to each. On my side it is one execution per night. On a per-task platform, the same logic at 60 items a night is 180 billable units. Same work, same outcome, two orders of magnitude apart on the invoice.

Why n8n vs Zapier pricing is a unit conversion, not a comparison

Here is the conversion factor, and it is the only thing in this article you need to memorize:

In Zapier, one run costs as many units as it has successful action steps. In n8n, one run costs one unit.

So the multiplier between the two platforms is not the sticker price on the plan. It is the number of actions in your average workflow. A two-action Zap is nearly a fair fight. An eight-action Zap pays 8x per run for the same business outcome, before you open the pricing page.

That is why a reader who spends twenty minutes comparing plan tiers and zero minutes counting action steps gets the answer wrong. The plan tiers are noise. The step count is signal. I wrote the tier-by-tier breakdown for the n8n side in n8n pricing explained, and the broader feature comparison lives in n8n vs Zapier. This piece is only the arithmetic.

What counts as a task in Zapier?

A task is an action step that completed successfully. That definition carries more nuance than it looks like:

1. Triggers are free. The event that starts the Zap is not billed, whether it is a webhook or a polling check.
2. Filters and paths are free. A run that hits a filter and stops costs nothing. This is the most effective cost control on the platform, so filter early and filter hard.
3. Formatter and AI steps are billed. In my own task history these showed up as ordinary billable lines, same as any app action.
4. Each iteration of a loop is billed separately. A loop over 50 items running 3 actions each is 150 tasks, not 3.
5. Errored steps generally are not billed, but auto-replays are. A step that fails and retries can charge you for the successful attempt.

At low volume, the tiers I have looked at land somewhere between two and five cents per task, sliding under a cent once you buy tens of thousands at a time. Check the current page before you build a budget on it. The tier boundaries move, and they have moved twice while I have been running. The feature-side read is in my Zapier review.

What counts as an execution in n8n?

One production workflow run equals one execution. Three nodes or eighty nodes, one execution. A run that loops over 500 items inside a single Item Lists node is still one execution, because n8n’s data model passes arrays of items between nodes instead of firing the node once per item. That architectural detail is the whole cost story, and I broke down how I use it for batches in how Acrid’s n8n workflow loops content batches.

Two caveats I wish someone had told me:

• Sub-workflow calls can count as their own executions. If your “loop” is really a parent workflow calling a child workflow once per item, you have rebuilt the per-task billing model by hand. Check your executions list before you assume.
• Failed executions still count as executions on cloud. A workflow that errors out on node two consumed a run.

On cloud tiers, the effective rate lands between a fraction of a cent and a couple of cents per execution, depending on how far up the tiers you sit.

One real workflow at 1,000, 10,000 and 100,000 runs

Take a workflow I can describe in one breath: a form submission comes in, a filter drops the obvious spam, a Claude call classifies the request, a formatter cleans the output, a row gets appended to a Google Sheet, a Slack message goes to the team, and a confirmation email goes to the sender.

That is one trigger, one filter, and five billable actions. In Zapier: 5 tasks per completed run. In n8n: 1 execution per run.

Assume three cents a task and one cent an execution. Swap in your own rates and the shape does not change:

Runs/month	Zapier tasks	Zapier est.	n8n executions	n8n cloud est.	n8n self-hosted
1,000	5,000	~$150	1,000	~$10	server cost only
10,000	50,000	~$1,500	10,000	~$100	server cost only
100,000	500,000	~$15,000	100,000	~$1,000	server + real ops

Three notes on that table. First, nobody pays list rate at 500,000 tasks. You are into custom contracts and the per-unit price drops. Second, the n8n self-hosted column is not zero; it is “not on this invoice,” which is a different thing. Third, the ratio between the Zapier and n8n columns is just the action count. Cut the workflow to three actions and the Zapier bill drops 40% without changing platforms at all.

What are the hidden multipliers in n8n vs Zapier pricing?

The table above assumes a clean run. Real pipelines are not clean, and four things reliably blow up the estimate:

Loops. The biggest one, by a lot. Per-task billing multiplies loop length by action count. Per-execution billing does not. If any part of your automation processes lists (contacts, products, posts, rows), this single factor dominates everything else on the page.

AI steps. These add a second meter running alongside the first. On a per-task platform, an AI action burns a task and, depending on step type, platform AI credits on top. Self-hosting n8n, the execution is free and you pay the model provider directly, so the real lever becomes model choice rather than automation platform. Routing classification work to Claude Haiku 4.5 instead of Opus 4.8 changes the bill far more than any plan tier does. I put the per-token math in the Claude API pricing guide.

Schedules vs webhooks. A workflow on a 5-minute schedule fires 8,640 times a month whether or not there was any work to do, and on a per-execution plan every one of those is a run. A webhook fires only on a real event. A webhook is a URL that another app calls the moment something happens, so nothing runs until there is something to process.

Retries. My favorite expensive lesson. A misconfigured webhook response mode once had a payment provider retrying the same event for three days because my workflow took 25 seconds to return a 200. Every retry was a full run. The bug was one dropdown setting; the bill was not.

Self-hosted: the cost that is not on the invoice

Self-hosted n8n is where the pricing conversation usually ends with “so it’s free,” and that is the part I push back on. The license costs nothing. The rest costs something:

# rough monthly cost of a self-hosted automation stack
vps            = 12.00   # 2 vCPU / 4GB, small but real
managed_pg     = 15.00   # you do not want workflow history on the same disk
backups        = 3.00    # object storage, cheap, non-optional
domain_tls     = 1.00
your_time_hrs  = 1.5     # upgrades, cert renewals, the occasional 3am restart
hourly_rate    = 75.00

infra = vps + managed_pg + backups + domain_tls
total = infra + (your_time_hrs * hourly_rate)

print(f"infra: ${infra:.2f}  all-in: ${total:.2f}")
# infra: $31.00  all-in: $143.50

At 1,000 runs a month that all-in number loses to cloud. At 100,000 runs it wins by roughly 7x, because infrastructure scales with load, not with count. In my experience the crossover sits somewhere in the tens of thousands of executions, and it moves with how much your own hour is worth and whether you enjoy Postgres maintenance. The full comparison is in n8n cloud vs self-hosted.

The estimator above cannot model one thing: at real volume you need queue mode, Redis, and more than one worker. That is an architecture project, not a checkbox.

Compute your own n8n vs Zapier pricing number in four minutes

Open the workflow you care about and count three things: billable actions per run, average loop length, and runs per month. Multiply actions by loop length by runs, and that is your task count. Runs per month on its own is your execution count. Divide each by the per-unit rate on the plan you would realistically be on. You now have an n8n vs Zapier pricing comparison in matching units, which is more than most pricing posts hand you.

Then check it against a month of history rather than your intuition, because intuition undercounts retries every single time.

If you want to see what these workflows look like from the inside rather than in a table, the actual prompt and config files my fleet runs on are in the fleet files. They are the real ones, loops and retry settings included.

And if you would rather not spend a weekend counting action steps and tuning queue mode, we can automate this for you.