Your Zapier AI Steps Probably Default to Premium. Audit Them Before December.

If you added an AI step to a Zap this year, there's a decent chance it's running on the most expensive tier without anyone choosing that. Zapier's help article on AI by Zapier model tier pricing, last updated October 1, 2026, says new AI by Zapier steps on paid plans default to Premium, which bills at five times the task rate. Q4 is when volumes go up for most of the businesses we work with, so this is a good month to look.

The rate card

From the help article, effective since June 15, 2026:

Tier	Task multiplier	Tool calls
Standard	1x	Not supported
Advanced	3x	Supported
Premium	5x	Supported
Bring Your Own Key	1x	Only with Advanced or Premium models

The formula: tasks used per run = (1 × model rate) + (number of tool calls × model rate). A Premium step that makes two tool calls costs 15 tasks. On Standard, with no tools, the same step is one task.

Separately, Zapier's pricing page gives "one MCP tool call = 2 tasks" as its illustrative rate, and Agents are billed in their own "activities" (400 a month on Free, 1,500 on the paid add-on) rather than tasks.

The default limit of 75 tasks per run pauses a step and asks for approval, which is a useful backstop and also a sign that the per-run cost can get there.

Why this is easy to miss

Zapier bills in tasks, and a plan is a bucket of tasks. A Professional plan at 2,000 tasks a month looks like 2,000 Zap runs. If 400 of those runs include a Premium AI step with one tool call, they consume 400 × 10 = 4,000 tasks on the AI step alone. You're over the plan before the other steps run, and the overage shows up as a plan upgrade rather than a line that says "AI."

We covered the general version of this in when to graduate from Zapier: the plan tier is a function of volume times steps, and the step count is where the surprises hide. AI steps with multipliers make the step count non-obvious.

The audit

Thirty minutes, in the Zapier editor:

1. List every Zap with an AI by Zapier step. Open each step and note the model tier. Anything created after June 15 on a paid plan is Premium unless someone changed it.
2. Ask what the step does. Classify, extract a field, rewrite a subject line, summarize? Those are Standard-tier jobs. Zapier's own table says Standard has no tool support, so the only reason to be on Advanced or Premium is that the step calls tools or needs the stronger model.
3. Switch and watch. Change the obvious ones to Standard. Keep a week of outputs side by side. In our experience with classify-and-extract steps, quality is rarely the thing that changes.
4. Count tool calls. For steps that stay on Advanced or Premium, look at how many tool calls a typical run makes. Each one multiplies.
5. Check the task history under your plan usage for the last three months, and compare against the Zap run count. The ratio tells you how much of your bill is AI multipliers.

Bring your own key

The BYOK row is interesting for anyone already paying an AI provider directly. At 1x, you pay Zapier one task for the step and your provider for the tokens. With the cheaper models now available (Claude Haiku 5.5 at $0.10 per million input tokens, which we wrote about in October), the token cost of a classification step is a fraction of a cent, well under the four extra tasks Premium charges. The trade is that you own the key, the quota and the provider relationship, which for a small team may not be worth the saved tasks.

When the answer is "move it"

If the audit shows one Zap consuming most of your tasks because of AI multipliers, that Zap is a candidate for a dedicated workflow. Our usual move is n8n, where a model call is a node and the cost is the tokens, not a multiplier. The comparison is in n8n vs Zapier vs Make. Zapier's pricing isn't unreasonable; it's just priced for breadth, and a single high-volume AI workflow is the opposite of breadth.

Either way, check the tier on your AI steps before the December volume hits. It's the cheapest fix on this list.