n8n vs Zapier in 2026: Pricing, AI Steps, and When to Move to Code

The real difference between n8n and Zapier is what each one counts. Zapier bills per task: every successful step in a workflow, including AI and code steps. n8n bills per execution: one full run, however many steps it has. So a 10-step workflow costs roughly 10 times more units on Zapier than on n8n. Zapier is easier to start with and has the bigger app catalog. n8n is cheaper at volume and can be self-hosted, so the data and the bill stay under your control. And both have a ceiling: past a certain size, a workflow is software, and is better written and tested as software. Below: 2026 prices, worked math, when each one wins, and the signs you have outgrown both. We build automations in code for clients, so read the last section with that in mind. We also tell people to use n8n when it fits.

---

2026 prices

Zapier, from zapier.com/pricing in October 2026:

What counts as a task: each successful action step, including AI by Zapier steps (priced by model tier), Code by Zapier runtime beyond the included amount, and Zapier MCP tool calls, all from one shared pool. Triggers, filters, paths and formatter steps don't count. With pay-per-task enabled, extra tasks cost 1.25 times the base rate on annual plans and 2.5 times on monthly plans, and usage pauses at three times your subscription.

n8n, from n8n.io/pricing in October 2026, billed annually:

An execution is one run of an entire workflow, "regardless of complexity". All plans include unlimited users and workflows, the AI Agent node, and an MCP server trigger and client. A standard self-hosted version is on GitHub.

The math on one real workflow

Say a workflow runs on every new order: look up the customer in the CRM, check stock, ask an AI step to classify the order, write a row to a spreadsheet, post to Slack, update the CRM. Six action steps. 2,000 orders a month.

• Zapier: 6 × 2,000 = 12,000 tasks, past the 10,000-task Pro tier ($129 a month billed annually). Expect that tier plus overage, and more if the AI step uses a premium model tier.
• n8n: 2,000 executions, inside Starter at €20 a month.

Double the steps and Zapier doubles. n8n doesn't move. That is the whole pricing story, and it is why multi-step and AI-heavy workflows tend to migrate.

When each one wins

Zapier wins when:

• you have a few short automations and want them working this afternoon;
• the apps you use are niche and Zapier is the one that supports them;
• nobody on the team wants to run a server, ever.

n8n wins when:

• workflows have many steps or run at volume;
• you call AI models inside workflows and don't want each call metered as a task;
• data has to stay on infrastructure you control, so you self-host;
• someone on the team is comfortable with a more technical tool.

Signs a workflow has outgrown both

A workflow tool is a great place for a process to start. These are the signs it should move to code:

• The canvas needs a guide. Forty nodes, nested branches, and one person who understands it.
• It writes to systems your business depends on, like the ERP, the CRM or invoicing, and a mistake costs money.
• Different people need different permissions, and the workflow runs with one admin connection that can see everything.
• You want tests before changing it, and there is no way to run them.
• Errors fail silently. A failed run is noticed when a customer complains.
• The bill grows faster than the work, because every step and every AI call is metered.

At that point the workflow is software, and it deserves version control, tests, error alerts and an owner. That is what we build: the same process, in code, connected to your systems with your users' own permissions, monitored, and maintained.

Related: what to expect from an AI automation agency, how much it costs to build an AI agent, and low-code, no-code or custom?

Also: OpenAI Agent Builder is shutting down on 30 November.