Zapier's AI steps and bringing your own API key are two different billing units. The built-in path spends tasks from your Zapier plan, at 1x to 5x depending on model tier. Bring your own key, and the model bill moves to a separate provider account in tokens, while the step drops to Zapier's cheapest task rate, scaling with tool calls.
Zapier AI steps vs bring your own API key: the short answer
Task cost and model cost are now two separate line items. Stick with Zapier's own AI account, and every run draws tasks from your plan, at a multiplier set by the model tier. Connect your own OpenAI or Anthropic account instead, and the model gets billed separately, in tokens, at your provider. Inside Zapier, that same step drops to the cheapest rate: one task per call, plus one per tool call, at 1x instead of 3x or 5x.
How Zapier's built-in AI steps actually bill you
Since 15 June 2026, AI by Zapier steps run on three pricing tiers, and each tier is a multiplier on Zapier's normal task unit, not a flat dollar figure. Standard runs at 1x tasks per run, and tools aren't available on this tier. Advanced runs at 3x, with tool access. Premium runs at 5x, with tools plus multi-step reasoning. As documented in August 2026, Standard runs lighter models like GPT-5 nano and Gemini 2.5 Flash, Advanced adds GPT-4.1 and Claude 4.5 Haiku, and Premium runs the heaviest, including GPT-5 and Claude 4.5-4.8 Sonnet and Opus. That line-up moves.
Which tier is the default for a brand-new step matters, since most people never touch it. Here's the catch: Zapier's own docs don't agree. The tier-pricing page, updated 20 August 2026, says Premium is the default. The announcement page, updated eight days earlier, says Advanced. Two live pages on the same help centre, disagreeing on the same fact. Don't take either page's word for it: open your own AI step and check which tier it's actually set to.
Zapier states the formula verbatim: “Tasks used per run = (1 x model rate) + (number of tool calls x model rate).” A tool call here is the AI step reaching out to a connected app or action mid-run, not the trigger that starts the Zap. Every tool call is charged at the same tier multiplier as the step itself, which is why a Premium step calling a few tools burns through tasks fast.
There's a safety net built in: if a single run of an AI by Zapier step hits 75 tasks, Zapier pauses it and asks you to approve before it continues. Treat that less as a hard ceiling and more as a signal that a Premium step with several tool calls got expensive enough to need a second look.
AI by Zapier steps live on Professional, Team, and Enterprise plans, not on Free (Free can preview one in an unpublished Zap, locked to Standard). If the rest of your bill looks off too, not just the AI steps, there's a full breakdown of how Zapier counts tasks that covers everything else.
How bringing your own API key bills you instead
So what changes if you route around Zapier's own AI account instead? Connect your own OpenAI or Anthropic account, and the model bill moves off Zapier entirely. Inside Zapier, that step now runs at the Standard rate: same formula, cheapest multiplier. Zapier's own worked example: a BYOK step making 4 tool calls costs (1 x 1) + (4 x 1), 5 tasks total. One catch: BYOK only works with Advanced or Premium models, you can't route a Standard-only model through your own key.
That 1x task cost inside Zapier isn't the whole bill. It's the cost of running the step. The model itself now runs on a second bill entirely, at your provider, denominated in tokens rather than tasks. A token is roughly a small chunk of text the model reads or writes, a few characters at a time. If you want the fuller picture, there's a full explainer on what a token actually is.
What BYOK actually asks of you, in practice: opening a developer account, generating and securing an API key, adding a payment method there, and checking a second usage dashboard that has nothing to do with your Zapier bill. Zapier's own docs call this “straightforward.” That's Zapier's characterization, not a neutral one. For a small operator already stretched on time to maintain new tools, a second billing relationship and a live API key inside a third-party workflow tool is real, ongoing overhead, not a one-time setup you check off and forget.
Zapier AI steps vs bring your own API key, side by side
Here's all of that laid out flat, so you're not doing the mental math mid-scroll.
| Path | What it bills | Who sends the bill | Task cost inside Zapier | Best for |
|---|---|---|---|---|
| AI by Zapier, Standard | Tasks, at 1x the model rate | Zapier | 1 task per run, no tool calls | Light, simple AI steps that don't need to call other apps mid-step |
| AI by Zapier, Advanced | Tasks, at 3x, plus 3x per tool call | Zapier | Variable, climbs with tool calls | Steps that need tool access but not the strongest model |
| AI by Zapier, Premium | Tasks, at 5x, plus 5x per tool call | Zapier | Variable, climbs fastest, can approach the 75-task pause | Steps that genuinely need the strongest model and several tool calls |
| Bring your own API key | Tokens, at your provider's own rates | OpenAI, Anthropic, or another provider, a second bill | 1x per call, the cheapest Zapier rate, still scales with tool calls | High-volume AI steps, or steps that would otherwise sit on Advanced or Premium with several tool calls |
Zapier's own formula, worked through on paper here, not a measured run:
- Standard, 0 tool calls: (1 x 1) = 1 task
- Advanced, 2 tool calls: (1 x 3) + (2 x 3) = 9 tasks
- Premium, 4 tool calls: (1 x 5) + (4 x 5) = 25 tasks
- Bring your own key, 4 tool calls: (1 x 1) + (4 x 1) = 5 tasks
Same shape of run, four different task counts, because the tier and the billing path move the number, not the work itself. For how tasks stack up against operations and credits on other platforms, there's a cross-vendor breakdown of pricing units worth a look.
Before you assume the built-in path is cheap: check which tier your existing AI steps are actually set to (Zapier's own pages disagree on the default), and count how many tool calls a typical run makes. Those two numbers, not your plan's sticker price, decide which path actually costs less.
When bring-your-own-key is worth the extra hassle, and when it isn't
There's no single right answer, and anyone who gives you one is probably selling something. Stay on Zapier's built-in account if your AI-step usage is light or occasional; if you'd rather manage one bill than open a second account, even at a modest extra cost; or if your steps only need Standard-tier models, where the built-in rate is already the cheapest one, so BYOK barely moves the number.
Switch to BYOK if your AI steps run often, especially on Advanced or Premium with several tool calls, and the task cost keeps climbing; if you're comfortable managing a provider account and a second dashboard without it becoming another chore; or if your automations regularly approach that 75-task pause, a concrete sign the built-in path is genuinely expensive at your volume. Still unsure which model fits the job? There's an honest picker for that too.
Don't rush either direction. For most small, occasional AI steps, built-in Standard is genuinely fine, and a second account to save a handful of tasks a month isn't worth the upkeep. The math only tips toward BYOK once volume or tool-call count actually climbs, and you can check that against your own account today rather than guessing. For the bigger picture on what AI automation costs a small team beyond this one toggle, that's worth a separate look.
Frequently asked questions
Should I give AI my API key?
Only to a tool you trust, with a key you can scope or rotate where the provider allows it. Treat a live API key sitting inside any workflow tool as a real security surface, not a set-and-forget step. Giving an automated step over-privileged access to a static key is a documented concern among people building with AI agents generally, not just inside Zapier.
What is Zapier AI best for?
Light, well-defined tasks inside an existing workflow: summarizing an inbound message, classifying a lead, drafting a reply, where the AI step is one part of a bigger Zap, not the whole job. It gets expensive fast once you stack tool calls on a higher tier, so it fits narrow steps better than open-ended reasoning work.
Is Zapier still relevant?
Yes, for teams that want a no-code way to connect apps without a developer, though it now competes with dedicated AI agent builders for more open-ended automation work.
Is there anything better than Zapier?
Depends what you need. For pure app-to-app automation, Make and n8n are the two most common alternatives, each with its own billing model. That's a platform-choice question though, separate from the one this article answers.
Does using my own OpenAI API key in Zapier still count as a Zapier task?
Yes. Connecting your own account doesn't make the step free inside Zapier, it drops it to the cheapest rate: 1 task per run plus 1 task per tool call. The model cost moves to your separate provider bill.
What happens if my Zapier AI step hits the 75-task limit mid-run?
Zapier pauses the step and asks you to approve it before continuing. It's a checkpoint, not a hard stop, and hitting it regularly is a real sign a run's tier and tool-call count are worth revisiting.
If you want to go deeper, there's a full explainer on tokens and how they differ from tasks, plus a diagnostic on the rest of your Zapier bill beyond AI steps. Worth a look before you flip that BYOK toggle. And if breakdowns like this one are useful, the newsletter covers more of them, sourced and dated, no vendor spin.