On June 15, 2026, Zapier quietly changed how every AI step in your automations gets billed. If you use "AI by Zapier" in any workflow — to classify leads, summarize emails, generate copy, route support tickets, or anything else — the math behind what each run costs is completely different now. Most people won't notice until they check their task usage at the end of the month and wonder why it doubled.

The shift is from flat-rate pricing to model-tier pricing. Every AI by Zapier step now has a tier — Standard, Advanced, or Premium — and each tier multiplies how many tasks a single run consumes. The default for new steps is Advanced, which costs 3x tasks per run. A Premium step with two tool calls burns 15 tasks in a single execution. That's a long way from the old flat rate of 1 task per run.

The tier breakdown

Zapier's pricing page lays out three tiers, plus a bring-your-own-key option. Here's what each one actually means for your budget.

Standard (1x tasks) runs the cheapest models: GPT-4.1 nano, GPT-5 nano, Gemini 2.5 Flash. No tool calls allowed. This is fine for simple classification, basic text generation, or anything where you just need a model to look at text and return a response. One run, one task.

Advanced (3x tasks) is the default for every new AI step you create. It unlocks tool calls — meaning the AI can trigger other Zapier actions during its run — and gives you access to GPT-4.1, o3-mini, Claude 4.5 Haiku, and Gemini 2.5 Pro. If you're doing anything where the AI needs to look up data in another app, write to a spreadsheet, or trigger a follow-up action, you're in this tier whether you asked for it or not.

Premium (5x tasks) is where the heavy hitters live: GPT-5, GPT-5.4, Claude 4.5 Sonnet, Claude 4.6 Opus, and Claude 4.8 Opus. These are the models that actually reason well, handle complex instructions, and produce output you don't need to rewrite. But five tasks per run adds up fast.

Your own API key (1x tasks) is the escape hatch. If you bring your own OpenAI or Anthropic credentials, you pay 1x tasks per run regardless of which model you use. You still pay the API provider directly for the tokens, but on the Zapier side, it's the cheapest option by far for anyone running AI steps at volume.

The math that matters

The formula is straightforward: tasks used = (1 × model rate) + (tool calls × model rate).

So a Premium step (5x) that makes two tool calls during its run uses (1 × 5) + (2 × 5) = 15 tasks. An Advanced step with one tool call uses 6 tasks. A Standard step with no tools uses 1 task.

If you're on a plan with 2,000 tasks per month and you have a workflow that runs 50 times a day with a Premium AI step that makes two tool calls, that's 750 tasks per day — 15 tasks per run times 50 runs. You'll burn through your entire monthly allowance in less than three days.

Compare that to the same workflow on Standard: 50 runs per day times 1 task per run equals 50 tasks per day. Same workflow, 15x cheaper.

The gotcha nobody warned you about

Here's what's going to bite people: Zapier defaults new AI steps to the Advanced tier. Every time you add a new AI by Zapier step to a workflow, it starts at 3x. If you don't manually change it to Standard, you're paying triple for what might be a simple prompt that doesn't need tool calls at all.

Legacy steps — anything created before June 15 — aren't automatically updated. They still run at the old flat rate. But there's a catch: as of June 30, legacy steps using knowledge sources with a Standard model have been paused. If your workflows stopped running in the last few weeks and you couldn't figure out why, this is probably it. You need to create a new AI by Zapier step and select Advanced or Premium to keep knowledge sources working.

There's also a 75-task safety net per run. If a single AI step hits 75 tasks in one execution, Zapier pauses it and asks you to approve before continuing. This protects you from a runaway loop where the AI keeps calling tools and burning tasks, but it also means your automation stops mid-execution until you manually intervene.

What to do right now

Three things, in order of urgency.

First, audit your existing AI steps. Open every workflow that uses AI by Zapier, check what tier each step is on, and calculate the real task cost per run. If a step is on Advanced but doesn't use tools, drop it to Standard. If it's on Premium and the task is simple classification, you're burning 5x tasks for no reason.

Second, check for paused legacy steps. If any of your automations stopped working around June 30, look for legacy AI steps using knowledge sources. Replace them with new steps on Advanced or Premium.

Third, consider bringing your own API key. If you run AI steps at any real volume — more than a few dozen runs per day — the BYOK option at 1x tasks plus direct API costs will almost certainly be cheaper than Zapier's tiered pricing. You'll need an OpenAI or Anthropic account, but the savings compound fast.

The bigger picture

This pricing change is Zapier catching up to reality. Running GPT-5 or Claude 4.8 Opus costs real money, and Zapier was absorbing that cost at a flat rate. The new model makes the economics transparent: cheaper models cost fewer tasks, expensive models cost more, and tool calls add to the bill.

The problem is transparency after the fact. If you built your automations six months ago and haven't looked at them since, you might be in for a surprise. The workflows still work. They just cost more than they used to.

Check your task usage now, before the end-of-month bill shows up.