Zapier vs Make vs n8n in 2026: pricing, reliability, and a real build

All three tools connect your apps and run workflows when something happens. The feature grids look similar. The bills do not. Each tool meters work differently, and that metering quietly decides what you can afford to automate. We verified every price below in June 2026 against the vendors' own pages and current third-party breakdowns. Prices change, so check before you buy. The logic of the comparison will hold longer than the numbers.

Three pricing models, three different bills

Zapier bills per task. A task is one successful action step. Triggers are free, and so are filters, paths, formatter, and delay steps. The free plan gives you 100 tasks a month, but only two-step Zaps, no webhooks, and 15-minute polling. There is no Starter plan anymore; Zapier retired it. The first paid plan is Professional at $19.99 a month on annual billing, or $29.99 billed monthly, for 750 tasks. A slider raises the cap: 2,000 tasks costs $49 a month on annual billing. If you go over, Zaps keep running and you pay around 1.25 times your base per-task rate.

Make bills per credit. Make renamed operations to credits in late 2025, and the idea is the same: every module run inside a scenario consumes one credit, including the trigger. Routers and filters are free. AI steps that use Make's own AI provider consume more than one credit. The free plan includes 1,000 credits a month. Core costs $9 a month on annual billing, around $11 billed monthly, for 10,000 credits. Pro is $16 and Teams is $29 a month on annual billing, both starting at 10,000 credits with sliders that scale into the millions.

n8n bills per execution. One complete run of a workflow counts once, whether it has 2 steps or 40. The cloud Starter plan costs 20 euros a month on annual billing, or 24 euros billed monthly, for 2,500 executions. Pro is 50 euros a month annual, 60 monthly, for 10,000 executions. Unused executions do not roll over. The Community edition is free to self-host, with no execution metering at all.

	Zapier	Make	n8n
Billing unit	Task: each action step counts. Triggers, filters, and paths are free	Credit: each module run counts, including the trigger. AI steps cost extra credits	Execution: one full run counts once, regardless of step count
Free plan	100 tasks, two-step Zaps only, no webhooks	1,000 credits, full scenario builder	Self-host the Community edition free, no metering
First paid plan	Professional: $19.99/mo annual ($29.99 monthly) for 750 tasks	Core: $9/mo annual (around $11 monthly) for 10,000 credits	Cloud Starter: 20 euros/mo annual (24 monthly) for 2,500 executions
10,000 of its unit	Around $130/mo billed monthly, around $90 annual	$9/mo on annual billing	Pro: 50 euros/mo on annual billing
Integrations	Around 8,000 apps, the largest catalog by far	Around 2,000 apps plus strong HTTP modules	Around 1,200 nodes, plus HTTP and code nodes that reach any API
Skill needed	None. Cleanest interface of the three	Moderate. Visual, but genuinely technical	Highest. Automation as code with a visual layer

Read that fourth row again. The units are not equivalent, and that is the point. On Zapier, 10,000 units might be 2,500 runs of a four-step workflow. On n8n, 10,000 units is 10,000 runs of any workflow. The more steps your workflows have, the more the execution model wins.

What self-hosted n8n actually costs

The software is free. The server is not, and your time is not. A small VPS that runs n8n fine costs around 5 euros a month at Hetzner or around $6 to $12 at DigitalOcean. A production-grade setup with Postgres and automated backups runs around $25 to $30 a month. Then comes the honest part: teams that track it report 1 to 3 hours a month on updates, backup checks, and debugging when a Docker image misbehaves. Price that time at whatever your hour is worth. At our $150 rate, that is $150 to $450 a month of labour on top of a $6 server. Self-hosting pays off at high volume or when a contract demands it. It does not pay off as a hobby.

Pick your tool by who is on your team

Feature lists do not make this decision. Your staffing does. Find your row.

No technical staff at all

Use Zapier. Full stop. The catalog of around 8,000 apps means your connector almost certainly exists, the interface is the cleanest in the category, and Copilot now builds working Zaps from a plain-language sentence. You will pay more per unit of work, and that is the correct trade: the alternative is automation that nobody on your team can fix on a Tuesday morning when it breaks. Do not let anyone talk you into self-hosted n8n. A server nobody maintains becomes the part of your business that fails silently.

One ops-curious person

Use Make. If someone on your team enjoys flowcharts, can read an API doc with effort, and gets a small thrill from a tidy spreadsheet, Make's scenario builder will fit them. Routers, iterators, and error routes handle logic that fights you in Zapier, at a fraction of the cost. The credit meter rewards efficient design, which suits exactly this kind of person. n8n Cloud is also workable here if your ops person is comfortable with JSON, but Make has the gentler slope.

A developer on staff, or on call

Use n8n. Code nodes in JavaScript or Python, raw HTTP against any API, branching on anything, and the option to self-host when data rules require it. One caution from our own projects: developer hours are usually a company's most expensive hours. If your developer is busy building product, a Make scenario your ops person owns may still beat an n8n instance your developer resents maintaining. Choose n8n because someone wants to own it, not because it is the most powerful option on paper.

The same automation, priced three ways

Abstract pricing models hide the real bill, so here is a concrete one. This is a standard lead-routing build, the kind of GTM integration work we set up most months. Four steps:

1. A form submission arrives by webhook.
2. Search the CRM for an existing contact or company.
3. Create or update the record and assign an owner by territory.
4. Notify the assigned rep in Slack.

Assume 2,000 leads a month. Same logic, same outcome on all three platforms. Here is the June 2026 math.

	Zapier	Make	n8n Cloud	n8n self-hosted
What gets billed	3 tasks per run (trigger and path are free; search, update, Slack each count)	4 credits per run (webhook, search, update, Slack; router is free)	1 execution per run	Nothing metered
Units per month	6,000 tasks	8,000 credits	2,000 executions	2,000 runs, unmetered
Plan needed	Professional, 10,000-task tier	Core, 10,000 credits	Starter, 2,500 executions	Community edition on a VPS
Monthly cost	Around $90 annual billing, around $130 monthly	$9 annual billing, around $11 monthly	20 euros annual billing, 24 monthly	Around $6 to $30 server, plus 1 to 3 hours of upkeep

Three honest readings of that table. First, Zapier costs around ten times Make for this exact shape. That is not a scandal; you are paying for the catalog and for the fact that anyone can maintain it. Second, Make is the price-performance winner for this workload, which is why it is our default recommendation for small teams with one capable operator. Third, self-hosted n8n is the most expensive option here once you price the maintenance hours honestly. At 2,000 runs a month, self-hosting saves you nothing. It starts winning at tens of thousands of runs, or the day a client contract mentions data residency.

What happens when a step fails

Every automation fails eventually. An API times out. Someone renames a CRM field. A rate limit trips at month end. The question is never whether a step will fail. It is what the platform does next, and whether anyone finds out before the pipeline runs dry.

Zapier: retry and alert

When a step errors, the run stops and you get an email. You can replay failed runs by hand from the Zap history. On Professional and above, Autoreplay retries a failed run automatically up to 5 times over about 10 hours, at 5 minutes, 30 minutes, 1 hour, 3 hours, and 6 hours. That schedule handles flaky APIs well. What Zapier does not give you is much room to respond inside the workflow: there is no clean way to say "if the CRM is down, write the lead to a holding table and page someone." Zaps that error repeatedly can also get paused automatically, which turns a noisy failure into a silent one if nobody reads the email.

Make: error routes you design

Make has the cleanest visual error handling of the three. Any module can have an error route attached, with directives that tell the scenario what to do: ignore the record and continue, resume with a substitute value, roll back, or break and store the run as an incomplete execution you can fix and resume later with the original data intact. The catch: none of this exists until you build it, and a scenario that keeps failing without error routes can deactivate itself. Wiring error routes before go-live is the single most skipped step we see in inherited Make accounts.

n8n: the most control, the most responsibility

n8n gives you per-node retry settings with configurable attempts and wait times, a continue-on-fail option per node, and a dedicated error workflow: a separate workflow that fires whenever anything fails, carrying the full context of which node broke and what data it held. You can also retry a failed execution from the exact node that failed, with the original payload. We typically wire the error workflow to Slack with the offending record attached, so a human sees every failure within minutes. Nothing is wired by default. n8n assumes an engineer is present.

One field note that applies to all three: the expensive failure is the silent one. A lead-routing automation that fails quietly for three weeks costs real revenue, and nobody budgets for it. Alerting on day one is non-negotiable on our builds. It is a core part of what we mean by RevOps hygiene.

AI steps in 2026: all three have them, maturity differs

By mid-2026, AI shows up in three layers on these platforms: assistants that build workflows for you, AI steps inside workflows, and agents where the model decides the steps. The three vendors are at very different depths.

	Zapier	Make	n8n
Build assistant	Copilot builds Zaps from plain language, included on paid plans	Maia builds scenarios from a description	AI Workflow Builder, with credits per plan (50 on Starter, 150 on Pro)
AI steps	AI by Zapier for classify, summarize, draft. AI fields in Tables. MCP tool calls cost 2 tasks each	AI Toolkit modules. Steps through Make's AI provider consume extra credits	Around 70 AI nodes built on LangChain, covering models, tools, and vector stores
Agents	Zapier Agents, a separate product with separate billing	Make AI Agents, in beta	AI Agent node: tool calling, memory across executions, RAG, human-approval steps
Model billing	Bundled into tasks and agent plans	Bundled credits, or bring your own API keys via app modules	Bring your own keys, pay the model provider at cost
Honest maturity read	Easiest to start. Linear: input in, output out	Middle. Watch the credit meter on AI steps	Deepest by a wide margin, and it is real engineering

Our honest read: most small businesses do not need an agent. They need one AI step inside a deterministic workflow. Classify this inbound email. Draft this reply for review. Summarize this call into the CRM. All three platforms handle that well today, and the cheapest place to run it is usually n8n with your own API key, because you pay the model provider directly instead of a marked-up bundle. If a vendor pitch hinges on the word "agents," ask one question: what exactly does the agent decide? If the answer is a flowchart, you need a workflow, not an agent. This is the boring, reliable end of AI build work, and boring is what holds up in production.

Security and compliance for client data

If your automations move client data, and lead routing does, the platform is part of your compliance story. Here is where each stands in mid-2026.

• Zapier holds SOC 2 Type II and complies with GDPR and CCPA. It runs on AWS in the United States, and Zap history, including payload data, sits on Zapier's servers for weeks. Fine for most SMB data. A problem the day a contract requires Canadian or EU data residency, because standard plans cannot offer it.
• Make is ISO 27001 certified, SOC 2 Type II audited, and GDPR compliant with standard contractual clauses for transfers. It operates both EU and US data centers, which gives you a residency option Zapier's standard plans lack.
• n8n Cloud is hosted in the EU on Azure, encrypts data at rest with AES-256, and aligns to SOC 2 with a public SOC 3 report. Enterprise customers can request the full SOC 2 report.
• n8n self-hosted is the only option where client data never leaves infrastructure you control. That sentence wins procurement reviews, and it comes with the matching duty: patching, TLS, backups, and access control are now your job. A neglected self-hosted box is less secure than any of the three clouds.

Four rules we apply on every client build, whatever the platform: send the minimum fields the workflow needs, keep one workspace per client, keep personal data out of logs and error messages, and rotate credentials when staff change. For Canadian non-profits, this is showing up in money terms: funders increasingly ask where donor data lives, and PIPEDA obligations follow the data. We cover this in our funding-readiness work for non-profits, because a clean answer on data handling is now part of winning grants.

The switching rules we use

• Your Zapier bill crosses around $300 a month: audit before you switch. Half the time the fix is consolidating wasteful Zaps, and you keep the tool your team already knows.
• A workflow needs loops, error branches, or waiting states, and building it feels like fighting the tool: that is a Make or n8n shape, not a Zapier shape.
• You are processing thousands of records on a schedule, enriching leads, or syncing systems continuously: n8n, almost without exception. The execution model exists for exactly this.
• Data residency appears in a contract or a funding agreement: self-hosted n8n is the only complete answer of the three. Make's EU region covers some cases.
• Migrations are rebuilds, not exports. Logic ports, configuration does not. Budget real hours per workflow, and pick for the next 18 months of volume, not the next 18 days.

When you do not need a consultant at all

You do not need us to connect a form to a spreadsheet. Zapier's Copilot and an hour of your time will do it, and the free plan may cover it entirely. The same goes for one-app notifications and simple digests. Hire help when the workflow touches revenue, when two or more systems have to agree on a single customer record, or when the cost of silent failure is bigger than the cost of the build. Our pricing is published: $150 an hour flat, scoped in writing first, no retainer minimums, and unused hours never expire. A typical lead-routing build like the one above is a few hours of work, not a project. And if a 30-minute call shows you can do it yourself, we will say so on the call. Book one if that would save you a quarter of dithering.

Frequently asked questions

Does Zapier still have a Starter plan in 2026?

No. Zapier retired the Starter tier. The lineup in June 2026 is Free (100 tasks a month, two-step Zaps only), Professional from $19.99 a month on annual billing for 750 tasks, Team from $69 a month on annual billing for 2,000 tasks and up to 25 users, and Enterprise on custom pricing. Triggers, filters, paths, and formatter steps cost zero tasks, which makes simple Zaps cheaper than the per-task model suggests.

Is self-hosted n8n actually free?

The software is free and unmetered. The realistic total is around $6 to $30 a month for a VPS with a proper database and backups, plus 1 to 3 hours a month of maintenance: updates, backup checks, and the occasional broken upgrade. Priced at a fair hourly rate, the labour is the real cost. It pays off at high volume or when data must stay on your servers, not for one small workflow.

Can I export my Zaps to Make or n8n later?

No. There is no clean export between any of these platforms. The logic ports because you understand it, but every workflow is a rebuild: new connections, new field mappings, new testing. Budget a few hours per workflow, and treat the migration as a chance to document everything, because the automation layer is usually the least documented part of a stack.

Which platform is best for AI in 2026?

For depth, n8n: around 70 AI nodes, agent loops with tools and memory, RAG with vector stores, and your own model keys so you pay providers at cost. For ease, Zapier: Copilot and AI steps anyone can use, with agents sold separately. Make sits between, and its bundled AI steps consume extra credits. Most businesses need one AI step in a fixed workflow, not an agent, and all three do that well.

What should a Canadian non-profit pick?

Usually Zapier's free or Professional tier, or Make's free 1,000 credits, which covers a surprising amount of donor and volunteer admin. Check your funding agreements first: if a funder or your privacy policy requires data to stay in Canada or the EU, Make's EU region or a properly maintained self-hosted n8n are the workable answers. Do not self-host unless someone specific owns the server.