Zapier is folding its standalone AI agents into the regular Zap editor — and pricing AI steps by model tier. Here's how the new task math works, and why it changes the Make vs Zapier calculation.
Zapier has been quietly restructuring how AI works inside its platform. Two moves matter for small businesses. First, AI by Zapier steps are now priced by model tier under a model-based pricing system Zapier rolled out starting June 15, 2026. Second, Zapier has published a migration guide moving standalone Zapier Agents into the core Zap editor — reporting around the change indicates the new AI by Zapier step with looping tool calls became generally available in mid-July. Instead of agents living in a separate product, AI reasoning becomes just another step inside a Zap.
Strategically, this makes sense. A Zap can now mix deterministic steps (filters, exact actions) with AI steps that handle judgment calls — classifying an email, drafting a reply, deciding which pipeline a lead belongs in. That's a more reliable architecture than a free-roaming agent.
The catch is the billing.
Zapier prices AI steps by model tier, each with a task multiplier:
| Tier | Multiplier | What it's for |
|---|---|---|
| Standard | 1x tasks | High-volume, simple prompts. No tool usage on this tier. |
| Advanced | 3x tasks | Stronger reasoning and tool usage. Default for new steps. |
| Premium | 5x tasks | Complex analysis, multi-step reasoning, heavy tool usage. |
The formula, per Zapier's help docs: tasks per run = (1 × model rate) + (number of tool calls × model rate).
Run the numbers on a realistic example. A Premium-tier AI step that makes four tool calls in a single run costs (1 × 5) + (4 × 5) = 25 tasks for one run. Run that Zap 40 times a day and you've consumed 1,000 tasks daily — from a single automation. Zapier does include a safety valve: if an AI step hits 75 tasks in one run, it pauses and asks for approval before continuing. That prevents runaway loops, but it doesn't make routine AI usage cheap.
One more wrinkle: legacy AI by Zapier steps created before June 15, 2026 aren't updated automatically and can't use tools. To get tool usage or pick a model tier, you have to build a new AI step — which lands you on the new pricing.
If your Zaps are simple and deterministic — form fills to CRM, Slack notifications, spreadsheet rows — nothing changes. Your task consumption stays flat and Zapier remains the easiest tool in the category.
If you've been building AI-heavy workflows, this is a "review now" situation:
We've said before that Make undercuts Zapier significantly for equivalent capability once workflows get complex. Model-tier multipliers widen that gap for AI-heavy use cases: Make bills per operation, and its pricing starts at $9/month, while an AI-heavy Zapier setup now consumes tasks at 3–5x rates plus per-tool-call charges.
That said, Zapier's counterargument is real. Building agent-style automation in the Zap editor — with approvals, run history, and guardrails built in — is far easier than assembling the same thing in Make or self-hosting n8n. For a non-technical owner who wants one AI workflow that just works, paying the multiplier may still beat spending a weekend on setup. The right answer depends on volume: at low run counts convenience wins, at high run counts the multipliers compound.
Zapier merging agents into Zaps is good product design and less generous billing, arriving in the same release. Audit your AI task consumption this week, downgrade model tiers where output quality holds up, and if your monthly task bill jumps, that's your cue to price out alternatives before you scale usage further.
See our full Make vs Zapier breakdown — pricing, power, and which one fits your business.
Read Make vs Zapier 2026 →