Zapier vs Make vs n8n 2026: The AI-Agent Pricing Trap
Zapier bills per task, Make per credit, n8n per execution. The same five-step workflow costs wildly different amounts depending on which word your invoice uses.
Every automation platform ships an AI agent now. Zapier has Agents, Make has Maia plus an agent builder, n8n put out a 2.0 release with a pile of AI nodes and real agent loops. The pitch on all three landing pages is the same: describe what you want in plain English and the platform assembles the workflow.
That part is mostly true. What none of them put on the pricing page is that the billing unit, not the feature list, decides whether you are paying $30 a month or $3,000 a month in a year.
Zapier bills per task. Make bills per credit. n8n bills per execution. Those sound interchangeable. They are not, and the gap widens with every step you add.
Prices below were checked against the vendor pages on 21 August 2026.
The short version
- Non-technical team, low volume, wants it to work without thinking: Zapier. The most expensive per unit, and the catalog and the polish are worth it under a few thousand runs a month.
- Ops-heavy, lots of branching, cost-sensitive: Make. Cheapest entry point, and the canvas handles complicated logic without turning into code.
- Dev team, high volume, or data that cannot leave your servers: n8n, self-hosted, where the per-execution cost stops mattering entirely.
The mechanic that decides everything
Zapier counts a task every time a single action step runs. A Zap that watches for an email, looks up a CRM record, and posts to Slack is three actions. Fire it a thousand times and you have spent three thousand tasks, not one thousand.
n8n counts one execution per workflow run, however many nodes are inside it. Thirty nodes, one execution.
Make sits in between. It counts a credit per module, so a five-module scenario costs about five credits per run, which puts it closer to Zapier’s shape than to n8n’s, at a much lower unit price.
A five-step workflow, ten thousand times a month
Trigger, two lookups, a filter, two outputs. Ten thousand runs.
On Zapier that is 50,000 tasks. Professional starts at $19.99 a month billed annually and includes 750 tasks, with everything above that billed per task, so 50,000 puts you somewhere in the low hundreds per month depending on which rung of the slider you land on. Run the slider yourself before you budget; I am not going to invent a number for you.
On Make that is roughly 50,000 credits. Core is $9 a month for 10,000 credits and scales from there, so call it a few tens of dollars.
On n8n Cloud that is 10,000 executions, which is the Pro plan at €50 a month billed annually. Flat, whether each run touches three nodes or thirty.
Same workflow, same volume, three different invoices, mostly because of how each vendor counts.
At a hundred thousand runs
This is where the surprise invoice arrives.
Zapier: half a million tasks, which is a custom contract and a conversation with sales.
Make: half a million credits, still linear but from a much lower base.
n8n Cloud: 40,000 executions is the Business plan at €667 a month, so a hundred thousand runs means Enterprise pricing.
n8n self-hosted: a hundred thousand executions on a $5 VPS. The Community Edition has no execution cap, so the count simply does not appear on any bill. That is the whole argument for high-volume teams. Self-hosted n8n does not get more expensive when you run more workflows; it gets more expensive when you run out of RAM, which is a cheaper problem.
A caveat so I am not lying to you. Self-hosting is not free the way a spreadsheet says it is free. Somebody patches it, backs it up, and gets paged when the queue jams at 2am. If that somebody bills at $150 an hour, the $700 a year you saved evaporates the first time it breaks on a holiday weekend. The math favors self-hosting at scale. It does not favor it for two people with no ops muscle.
One more thing worth knowing before you build a business on it: Community Edition is fair-code, not open source. Self-hosting it to run your own company’s automations is squarely allowed. Reselling it as your own hosted service is not.
What the agents actually do
The pricing is the decisive part. The agents are why you are reading this in 2026.
Zapier Agents ride on the widest app catalog of the three by a distance. Give an agent a goal and access to your connected apps and it decides which actions to take. Reach is the strength. The weakness is that every action it takes is still a task, and an agent that fumbles through five tools to answer one question has just spent five of them.
That got sharper in June. Since 15 June 2026, AI by Zapier steps are billed by model tier: standard counts as one task, advanced as three, premium as five, and new steps default to advanced. Connect your own AI account and you are back to 1x. There is also a safety valve that pauses a Zap if a single run burns 75 tasks, which tells you what they were seeing in the wild. Autonomy and per-task billing remain an uncomfortable pairing, and now the multiplier is explicit.
Make split it in two. Maia is the conversational builder that assembles modules on the canvas from a description. Make AI Agents live in that same canvas with a reasoning panel that shows each decision as it happens, which is genuinely useful, because when an agent does something stupid you can see why instead of guessing. It is still labeled beta. AI modules also eat more than one credit each, so agent-heavy scenarios burn faster than the sticker rate suggests.
n8n 2.0 is the one built for people who want to ship an agent rather than demo one. LangChain integration, around seventy AI nodes, agent memory that persists across executions, vector store support for retrieval, sandboxed code execution, MCP support for external tool servers, and real loops where the model calls a tool, reads the result, and iterates. It is the most capable and the most complex. If your team can read a LangChain diagram it gives you the most headroom. If they cannot, the other two get you to a working agent faster.
None of them has the best agent in the abstract. Zapier bet on breadth of connection, Make on visibility, n8n on depth.
The compliance question skips the math entirely
If your data cannot leave your infrastructure, this section is the decision and you can ignore everything above.
Zapier and Make are cloud only. Your payloads transit their servers. You can sign a DPA and be satisfied with their posture, and for plenty of teams that is fine.
n8n Community Edition runs on your box. Workflow data, credentials, payloads, none of it needs to touch n8n’s cloud. For a regulated team that is not a nice-to-have, it is the difference between approved and no. This is why engineering organizations pick n8n even when Make would be cheaper to run.
The snapshot
| Platform | Entry | Billing unit |
|---|---|---|
| Zapier | Free 100 tasks; Professional $19.99/mo annual for 750 tasks, overage per task | Per task, meaning per action step |
| Make | Free 1,000 credits; Core $9/mo for 10,000 credits | Per credit, meaning per module, and AI modules cost more |
| n8n Cloud | Starter €20/mo for 2,500 executions; Pro €50/mo for 10,000 | Per execution, meaning per workflow run |
| n8n self-hosted | Free software, $5 to $7/mo VPS | None. You pay for the box |
Two footnotes. Make moved from operations to credits in 2025 and the AI modules do not cost one credit each, so an AI-heavy scenario costs more than the flat count implies. And n8n’s free tier is self-hosting, not a free cloud plan.
Switching later is the expensive part
Whatever you pick, the workflow logic gets built inside that platform’s model. Zaps, scenarios, nodes. None of it exports cleanly to the others. A hundred production workflows on Zapier is a hundred workflows you rebuild by hand when you migrate to escape the task bill.
So the real cost is not this month’s invoice. It is that the platform you choose at 10k runs a month is the one you are stuck with, or expensively unstuck from, at 500k. Which flips the question. Do not optimize for what is cheapest today.
If you are genuinely going to stay small, Zapier’s premium buys back time you would otherwise spend fiddling, and that is a fair trade. If you can see six figures of monthly runs coming, start on n8n now, while the migration you are avoiding is still hypothetical.
The cheap experiment, if you are on the fence: build your single most important workflow on Make’s free tier and on a self-hosted n8n at the same time, run each a few hundred times, and read both invoices. The billing model stops being abstract the moment one of them charges you five credits for what the other counted as one run.