What an AI Step Actually Costs in Zapier Tasks

A Zapier AI step on the current default tier costs 15 tasks a run, not one. The formula, the 3x-to-5x default change, and what both do to a plan.

A single pale cream square on a dark teal field, set against a tightly packed block of fifteen identical squares beside an amber vertical rule
12 min read
2,486 words

A Zap that has stopped is easy to diagnose. A Zap that runs perfectly and quietly empties the month is not, and Zapier AI step task usage is where most of that money goes. On the Professional plan at the 750-tasks-a-month tier, one AI step on the tier new steps are created with today, calling two tools, costs 15 tasks per run. Fifty runs spends the entire plan. The step still works. The invoice is the only place anything looks wrong.

The arithmetic behind that is published, which is the frustrating part. It is published across three different pages, none of which is the pricing page most people read, and the single number that matters most — which tier a new step is built on — is not the same on all of them.

A user posting as dvano raised exactly this on Zapier’s community forum: an eight-step Zap burning 48 tasks when they counted 14 at most. That is one person’s report rather than evidence of a systemic problem, and the thread closed without the discrepancy being pinned down. What makes it worth reading is SamB, a community manager at Zapier, pointing at the thing almost nobody counts — that every successful call to a knowledge source is billed like a tool call, at the step’s own rate.

What Zapier actually charges you for

A task is a successful action. That is the whole definition, as Zapier’s task-usage documentation gives it, and most of the usefulness lives in what it excludes: triggers never cost anything, however often they poll; filters and paths are free; steps that error or halt are free; testing in the Zap editor is free even though the action genuinely fires in the connected app.

The built-in apps are free too, and the list is longer than people expect — Formatter, Delay, Looping, Sub-Zap, Digest, Zapier Manager, Storage, Tables and Forms. A Zap that fetches nothing and writes nowhere, however many steps of logic it runs, costs nothing to run.

Against that, three products bill above the standard rate. Zapier MCP charges 2 tasks per successful tool call. Lead Router charges 5 tasks per lead routed. Code by Zapier charges 1 task per run, plus another for every 30 seconds beyond the plan’s runtime allowance. And one structural detail from the pricing page is worth holding onto before any of the maths below: tasks are a single shared pool. Zaps, AI steps, code, MCP and the SDK all draw from the same allowance, so there is no separate AI budget to blow through first.

The Zapier AI step task usage formula

Since 15 June 2026, AI by Zapier has been priced by model tier rather than per step. Zapier publishes the formula directly:

Tasks used per run = (1 × model rate) + (number of tool calls × model rate)

The model rate is the multiplier for the tier the step is set to. Standard is 1x and supports no tools at all. Advanced is 3x. Premium is 5x. Connecting your own model provider — Bring Your Own Key — is 1x, and available with Advanced or Premium models only.

The detail that does the damage is that the multiplier applies twice over. It sets the base cost of the step, and it is charged again for every tool call the step makes. A Premium step that calls two tools is not 5 tasks plus 2. It is (1 × 5) + (2 × 5), which is 15. The same step on Standard, if it needed no tools, would be 1.

And Zapier defines a tool call broadly: any time the step successfully uses a tool, meaning an app action or a knowledge source. Attach three knowledge sources to a Premium step and let the model consult all of them, and you have bought 20 tasks for one run without adding a single visible step to the Zap. This is where a total that “makes no sense” usually comes from, and it is why the model’s own autonomy is a billing question — the agent decides at runtime which tools to use, so the cost of a run is not fixed when you build it.

Zapier help centre table listing Standard at 1x, Advanced at 3x, Premium at 5x and Bring Your Own Key at 1x, with a note that new steps default to Premium
Zapier's model tier table, captured 21 Nov 2026: Premium is 5x and is the default for new steps.

The default moved, and the old page still reads as current

Zapier’s product-update note announcing model-based pricing is unambiguous about what a new step would be created with. Under the heading “What does this mean for you?”, it says: “The default tier is Advanced: New steps default to the Advanced tier (3x) to allow you to use tools with your AI by Zapier step.”

The live model tier pricing page, updated on 20 November 2026, says something else. Premium’s row in the table is annotated “The default for new steps”, and the note beneath it reads: “When you create a new AI by Zapier step on a paid plan, the model tier defaults to Premium.”

Both pages are current, both are Zapier’s, and they disagree. The June page carries a small standing disclaimer that its information was accurate at the time of publication, which is the only signal that it is no longer the one to trust — and it is not a signal anyone reads a help page looking for. The practical consequence is that a step built in the summer and a step built this week start on different tiers, and a step duplicated from an old Zap does not necessarily match a step built fresh beside it. The difference between 3x and 5x on a step with two tool calls is 9 tasks against 15, every run, forever, for identical output.

Working out the bill before you build

The calculation is short enough to do in your head, and worth doing before the step exists rather than after the month ends.

Pick the tier deliberately. It is a dropdown on the step, and leaving it alone is a decision with a 5x price on it. Classification, extraction, short summaries and routing rarely need a premium model; those are Standard work, at 1x and no tools.

Count every tool, then count the knowledge sources too. Both are tool calls. If the step can reach five tools and the model is free to pick, budget for the worst case, not the typical one.

Apply the formula, then multiply by expected runs. This is where an automation that fires per inbound email differs from one that fires per week.

Hold the result against the plan allowance. Here are the real numbers, in USD, taken from Zapier’s pricing page with its currency control set explicitly to USD. Free gives 100 tasks a month. Professional at the 750 tasks a month tier is $29.99 a month billed monthly, or $19.99 a month billed yearly — and the page opens on the yearly figure, labelled as saving 33%, so the $19.99 most people quote is the annual-commitment price. Team starts at 2,000 tasks for $103.50 a month billed monthly, $69 billed yearly. Enterprise is quote-only.

At $29.99 for 750 tasks, a task costs almost exactly four cents. The 15-task Premium step is therefore about 60 cents a run, and 50 runs a month is the whole plan. The same workflow on Standard, if it can be done without tools, is 1 task a run: 750 runs on the same money. That is a fifteenfold difference in what a plan buys, decided by a dropdown.

Zapier pricing page with the currency control set to USD, the 750 tasks per month tier selected and monthly billing chosen, showing Free at $0 and Professional at $29.99 per month
Zapier pricing in USD at the 750 tasks/month tier, billed monthly, captured 21 Nov 2026.

Loops multiply everything downstream

Looping by Zapier is free. That is the trap rather than the saving, because of what Zapier’s own documentation says sits beneath it: “All actions after the looping step run for each iteration of the loop.”

Not the actions inside some visually nested block — everything after it. A loop over 200 line items with a Premium AI step somewhere below it does not cost 15 tasks. It costs 3,000, which is four times the Professional allowance in a single run. The looping step itself contributes nothing to that figure, which is precisely why the cost is hard to see when you are reading the Zap top to bottom.

Zapier documents the fix, and it is a filter rather than a setting. After the last step you want to repeat, add a filter that continues only when loop_iteration_is_last matches Is True. The filter runs on every iteration and costs nothing, but only passes on the final one, so any action beyond it runs once and uses 1 task. It is the difference between a summary step that fires 200 times and one that fires when the loop is done.

Four ways to cut the number without cutting the workflow

Drop to Standard where the model is doing clerical work. Tier is per step, not per Zap, so a classification step can sit at 1x in the same workflow as a drafting step at 5x. Most AI ticket triage workflows are sorting and labelling, which is the cheapest thing a model does.

Bring your own key. Connecting your own provider account drops the step to 1x and the tool calls with it, and you pay the model provider directly for inference. For a tool-heavy step this is the largest single reduction available — Zapier itself calls it the most cost-effective option for tool-heavy workflows. The trade is that you are now managing a second bill and a second set of rate limits, the same calculus that governs whether to self-host a coding model at all.

Set the per-run task limit on purpose. Every AI step carries a ceiling on how many tasks one run may use. The default is 75, and it can be raised as far as 500. Hit it during a live run and Zapier pauses the step and asks you to approve before it continues. Read it correctly: 75 is the ceiling on one run, not on the month. On a Premium step that is fourteen tool calls in a single execution, so the default only interrupts a step that genuinely queries a lot of resources — and raising it toward 500 removes the one automatic brake in the system.

Prune the knowledge sources. Each successful call to one is billed at the step’s rate, so an attached source that is rarely useful is a recurring charge for a rare benefit. The same discipline applies to any AI job that runs at volume rather than on demand, which is the lesson from what bulk alt-text generation costs once the volume is real.

What changes on 1 December 2026

Next Gen Zaps and the Zapier SDK have been free. That ends on 1 December 2026 at 12pm ET — 17:00 UTC — and Zapier is explicit that the free period ends for all customers at once rather than rolling out gradually.

From that moment each app action and each external call in a Next Gen Zap costs 1 task, as does each inbox-lease trigger in its code. The trigger that starts the Zap stays free. SDK calls made outside Zapier entirely, on your own servers or in a cron job, cost 1 task each, and a batch or bulk call is charged per item in the batch rather than per call. Classic Zaps are mostly untouched; the exception is a Classic Zap making SDK calls inside a Code by Zapier step, which pays the step’s base task and then 1 task per SDK call on top.

Two changes land the same day that cut the other way. Free By Zapier apps — Tables, Forms, Storage — become free everywhere, including through MCP and the SDK, where some previously attracted the MCP rate. And AI by Zapier gets its own price everywhere, so a step reached through MCP is billed at its tier rate rather than the flat MCP rate of 2.

One operational gap is worth knowing before the date rather than after it: per-run task counts for Next Gen Zaps are not available yet. Zapier currently offers account-level totals and billing settings, with task history listed as coming later. If usage jumps in December, the tooling to attribute the jump to a specific run does not exist yet.

What the pricing page will not tell you

The public task usage rates page still lists the SDK as “Free in beta”, with a footnote promising notice before the beta ends. The help centre has already given that date. A reader who checks the page most likely to be linked from a sales conversation gets the older answer.

The pay-per-task rate is not published anywhere. Zapier’s billing documentation says the rate depends on your plan and billing cycle and tells you to check Billing settings for the exact figure — so a number quoted elsewhere is somebody’s account, not a rate card. What is published is the ceiling: pay-per-task usage stops at 3x your plan’s task limit, after which runs are held until the cycle resets or you upgrade. Budget against that cap, which is knowable, rather than against a rate that is not. Note also that pay-per-task is on by default for accounts created after January 2024, and on Professional and Team only the account owner can switch it off.

Legacy AI steps behave differently again. Steps created before 15 June 2026 were not migrated, cannot use tools or knowledge sources, and will not show a tier selector — and since 30 June 2026, legacy steps using knowledge sources on a Standard model have been paused outright. Getting tools means building a new step, not editing the old one.

Finally, Zapier states on the rates page itself that rates may change as it introduces new capabilities or updates how tasks are measured, with a promise to announce changes before they apply. The Advanced-to-Premium default sitting on two pages at once is a reasonable argument for reading that promise as covering the rates rather than every default around them. The cost of an automation is not a thing you establish once. It is a thing that drifts, in the same quiet way a Gmail send pipeline drifts into its daily quota — and the number worth writing down somewhere is not the price, but the date you last checked it.

Sources

Every figure in this article traces to one of these. Dates are when the page was read.

  1. AI by Zapier model tier pricing
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  2. How is task usage measured in Zapier?
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  3. AI by Zapier: new model-based pricing starting June 15, 2026
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  4. Task usage rates
    ZapierAccessed Nov 21, 2026Vendor pricing
  5. How Next Gen Zaps and the Zapier SDK are billed
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  6. Plans & Pricing
    ZapierAccessed Nov 21, 2026Vendor pricing
  7. Loop your Zap actions
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  8. How pay-per-task billing works in Zapier
    Zapier Help CenterAccessed Nov 21, 2026Official docs
  9. Issues with task consumption in Zapier automation steps
    Zapier CommunityAccessed Nov 21, 2026Vendor figure

Written by

Kai Devlin AI Automation Engineer

Full-stack developer who builds and runs AI automation systems in production. Runs local LLMs on personal hardware, builds N8N pipelines that actually ship, and deploys on Cloudflare Pages. Every guide on Pipeline Monk is tested on real consumer hardware — a Ryzen 7 5800HS with 16GB RAM and a GTX 1650. If it works on that, it works.