How much does Zapier cost and is it worth the price in 2026?

Zapier bills on tasks, and as of July 2026 it had not moved to the credit or activity billing some competitors adopted. A task is counted each time a Zap completes an action that moves data. Triggers, polling, filters, paths, and built-in tools such as Tables, Formatter, Filter, Path, and Delay do not consume tasks — which is the single most useful fact for keeping a Zapier bill predictable.

Zapier plans (as of July 2026)

Task allowances scale with plan size, and Zapier's product line — Zaps, Tables, Interfaces, Chatbots, Canvas, and Agents — carries per-product allowances that differ by plan, with some unlimited on a given tier and others capped. Those allowances change often enough that Zapier's official pricing page is the authoritative reference for current limits. Over-limit task usage is billed automatically at roughly 1.25 times the base task rate, so Zaps keep running rather than halting.

Why task placement matters more than task price

Because only data-moving action steps count, the order of steps in a Zap determines cost. A filter placed after an action pays for that action on records it then discards; the same filter placed first does not. This is the lever that most changes a real Zapier bill.

How Zapier compares

Zapier's task model is the easiest in the category to forecast, and its 9,000-plus app catalogue is the broadest. It typically loses on price-per-volume to Make's credits or self-hosted n8n, and on logic depth to code-first Pipedream, while winning on integration coverage and speed to a working automation.

Editor's Note: Across ShadowGen client builds on Zapier in 2024-26, the cost surprises are almost always about task placement, not task price. On one DTC retail client, multi-step Zaps were burning far more tasks than forecast because a filter sat after two action steps instead of before them, so the Zap paid for actions on records it then threw away. Moving the filter to the first step cut that client's monthly task consumption by roughly 30% with no change in output. — Rafal Fila, ShadowGen