Zapier Review 2026: AI Step Pricing Changed, Does the Verdict?
Published June 28, 2026 · Updated June 28, 2026 · by Pondero Reviews
The short version
Zapier's June 15 AI model-tier pricing makes a default agentic step burn 9 tasks per run, not 1. We pulled the current pricing and worked the math to see if the verdict still holds.
Pros
- ✓Roughly 8,000 app connectors, still the widest integration graph in automation by a wide margin
- ✓The linear Zap builder is the lowest learning curve in the category for a non-technical team
- ✓Zapier Agents is GA, adding a chat-driven assistant layer on top of the Zap you already built
- ✓A 75-task per-step circuit breaker pauses runaway AI steps before they drain your task balance
- ✓Bring-your-own-key keeps AI steps at the 1x multiplier, so tool-heavy agents stay affordable if you hold the API account
Cons
- ✕AI steps now multiply by model tier, and the default Advanced tier turns a one-task step into nine with two tool calls
- ✕At Professional's 750 tasks, a single Advanced AI step with two tool calls leaves room for about 83 runs a month
- ✕Zapier Functions is being deprecated, so the web code IDE we listed as a differentiator is going away by September 1
- ✕The Free plan's 100 tasks a month is still too thin to validate any real AI workflow
- ✕Tool calls in AI steps are not available on Enterprise, an odd gap at the top tier
Zapier Review 2026: AI Step Pricing Changed, Does the Verdict?
Short answer: the verdict mostly holds, but the rating slips to 4.1 from 4.3, and the reason is a pricing change you may not have noticed. As of June 15, 2026, AI by Zapier steps are no longer a flat one-task cost. They multiply by model tier, and the default tier for any step that uses tools is Advanced, which bills 3x (Zapier help, June 15 2026).
Here is the number to leave with. A single AI step on the default Advanced tier with two tool calls burns 9 tasks per run, not 1. On the Professional plan, 750 tasks a month, that is about 83 runs of that one agentic Zap before you upgrade (Zapier pricing, pulled June 28 2026). The math is the whole story of this refresh, so we work it out below.
Zapier is still the right call when connector breadth and the shortest path to a working automation outweigh cost. What changed is the cost shape on the AI side, and for one kind of buyer it changes the buy decision outright.
What Zapier is, and who it's for
Zapier connects roughly 8,000 apps through a linear, no-code workflow builder. You pick a trigger, chain actions, and Zapier fires the chain when the trigger event lands. On top of that sit AI steps, Chatbots, and now Zapier Agents, a chat-driven assistant layer. The platform's whole pitch is breadth and ease: a non-technical ops person can wire two SaaS tools together in an afternoon without a developer.
That fit has not changed. If your team is non-technical and you need to connect a lot of apps fast, Zapier is built for you. The question this review answers is narrower: now that AI steps cost more, does that fit still hold for AI-heavy work? For most readers the answer turns on volume, and the formula below is how you find your own break point.
The AI step pricing change, worked out
This is the reason to read a June review instead of an April one. Before June 15, an AI by Zapier step cost one task per run, flat. Now the cost depends on which model tier you pick, and every tier above Standard multiplies the base run plus each tool call.
The tiers, straight from Zapier's pricing doc (Zapier help, model tier pricing):
| Model tier | Task multiplier | Tool support | Notes |
|---|---|---|---|
| Standard | 1x | No | Simple prompts, no tools |
| Advanced | 3x | Yes | Default for new steps with tools |
| Premium | 5x | Yes | Best for complex tool workflows |
| Bring your own key | 1x | Yes | Your own API account, keeps the 1x rate with tools |
The formula Zapier publishes is short:
Tasks per run = (1 x model rate) + (number of tool calls x model rate)The first term is the step running itself. The second is every tool the AI calls during the run, where a tool call is any successful app action or knowledge-source lookup (Zapier help, model tier pricing). Run it for two realistic cases.
A simple classification step, say "read this support email and tag it urgent or not," uses no tools. On Standard that is (1 x 1) + (0 x 1) = 1 task. Same as before the change. Nothing to see here, and if all your AI steps look like this, the new pricing costs you nothing.
Now an agentic step, the kind people actually build in 2026: an AI step that reads a CRM record, enriches it, and writes the result back. That is three tool calls, on the default Advanced tier. The math: (1 x 3) + (3 x 3) = 12 tasks per run. One run of that Zap eats 12 of your monthly tasks.
Here is what each shape costs per run, with the monthly run ceiling at Professional's 750 tasks:
| AI step shape | Tier | Tool calls | Tasks per run | Runs per month on 750 tasks |
|---|---|---|---|---|
| Classify, no tools | Standard (1x) | 0 | 1 | 750 |
| Classify, no tools | Advanced (3x) | 0 | 3 | 250 |
| Enrich, light | Advanced (3x) | 2 | 9 | ~83 |
| Enrich, agentic | Advanced (3x) | 3 | 12 | ~62 |
| Complex reasoning | Premium (5x) | 2 | 15 | 50 |
Zapier's own worked examples confirm these rates: a Premium step with two tool calls is (1 x 5) + (2 x 5) = 15 tasks, and an Advanced step with one tool call is (1 x 3) + (1 x 3) = 6 tasks (Zapier help, model tier pricing).
The jump from 750 runs to 62 is the headline. If you built an agentic Zap in May and it ran fine inside your task budget, the June 15 change can cut your effective run ceiling by an order of magnitude without changing a single line of your workflow. Worth a calendar reminder to check your task usage.
Two guardrails soften the blow. A per-step circuit breaker pauses the Zap and asks for manual approval if any single step hits 75 tasks in one run, so a misconfigured agent in a loop can't silently drain your balance (Zapier help, June 15 2026). And bring-your-own-key keeps the multiplier at 1x while still allowing tools, so if you hold the OpenAI or Anthropic account, that same three-tool agent costs (1 x 1) + (3 x 1) = 4 tasks instead of 12 (Zapier help, model tier pricing). For anyone running AI steps at volume, BYOK is the single biggest lever on the bill.
The change is available on Professional, Team, and Enterprise, not Free. One odd gap: tool calls in AI steps are not available on Enterprise accounts at all, though Enterprise users can still run AI steps without tools (Zapier help, June 15 2026).
Zapier Functions is being deprecated
The web-based code IDE that let you write standalone JavaScript and Python is on its way out. Zapier stopped accepting new Functions users on June 1, 2026, existing functions keep running until September 1, 2026, and after that date they stop executing and the interface goes away (Zapier help, Functions deprecation).
Our April review listed Functions as a differentiator. That now misleads, so treat this as the correction. The replacement is Code by Zapier, which runs JavaScript or Python natively inside a Zap and supports npm and PyPI packages plus AI-assisted code generation (Zapier help, Functions deprecation). If you relied on Functions for standalone logic, your migration path is a Code by Zapier step. If you wanted code as a first-class building block rather than a step inside a Zap, this is a good moment to look at Make's Code App or n8n's Code node instead.
Current pricing and the task math
What you pay, current per Zapier's pricing page, pulled June 28 2026. Annual figures come from the June 28 pricing-page fetch:
| Plan | Billed monthly | Billed annually | Tasks/mo |
|---|---|---|---|
| Free | 0 USD | 0 USD | 100 |
| Professional | 19.99 USD/mo | 14.99 USD/mo | 750 |
| Team | 69 USD/mo | 52 USD/mo | 2,000 |
| Enterprise | Custom | Custom | Custom |
A task is one action Zapier completes. A two-step Zap that fires 100 times a month uses 200 tasks, because the trigger is free but each action counts. That is the part newcomers miss: your bill scales with your step count times your run count, not just your run count.
Layer the AI change on top and the estimate gets sharper. If your Zaps are classic SaaS-to-SaaS plumbing with no AI steps, the old task math holds and Professional's 750 tasks goes a long way. The moment you add a default Advanced AI step with a couple of tool calls, divide your remaining task budget by 9 or 12 to get your real run ceiling. When that number drops below what you need, your options are BYOK to cut the multiplier, upgrade to Team's 2,000 tasks, or move the AI-heavy workflow to a platform that doesn't meter reasoning loops this way.
Zapier Agents
Zapier Agents is generally available, a chat-driven assistant that sits on top of your Zap workflows. You describe what you want in natural language and the agent assembles or triggers automations, rather than you dragging steps in the linear builder. It is the same shift the whole category is making toward conversational orchestration.
Against the competition, the shape differs. Make's AI Agents went GA on all paid plans and, as of June 2026, can call external MCP tools and native modules in one reasoning loop with the model picking the tool (Make community, June 4 2026). n8n's AI Agent node is the most battle-tested implementation and runs on your own server. Zapier Agents leans on Zapier's connector breadth, which no one else matches, but the autonomous tool-selection loop is more mature on Make and n8n today. If the agent layer is your reason for buying, that gap matters.
What Zapier still does better
Connector breadth is the moat, and it has not moved. Roughly 8,000 integrations versus Make's 3,000-plus and n8n's smaller native catalog means the obscure app your finance team depends on is far more likely to have a prebuilt Zapier connector than a prebuilt anything-else connector. For a team whose automation list is long and weird, that breadth is worth paying for on its own.
The linear builder is the second edge. There is no spatial canvas to learn, no node graph, no Docker. Trigger, action, action, done. For a non-technical operator who wants one workflow live today, that is the shortest path in the category, and the AI pricing change does nothing to dull it.
Reliability rounds it out. Zapier's platform-side track record on actually firing your Zaps when the trigger lands is strong, and a managed cloud means you never patch a server or babysit an upgrade. That last point is the quiet reason a lot of teams stay even when the raw dollar math favors a self-hosted option.
What got harder to justify
AI steps at volume are now a real line item. A 3x to 5x multiplier on agentic runs is not a rounding error once you are running an AI Zap hundreds of times a month, and the default Advanced tier means you opt into the higher cost unless you actively switch. BYOK or Standard fixes it, but you have to know to reach for them.
Functions going away removes a feature we previously credited. Code by Zapier covers most of the gap, but anyone who built on the standalone IDE has a migration on the calendar before September 1.
The Free plan stays too thin to matter. At 100 tasks a month, an AI step on the Advanced tier with two tool calls burns through your entire monthly budget in 11 runs. Free is a demo, not a tier you can validate a real AI workflow on, and the new pricing makes that gap wider.
Zapier vs Make vs n8n, in brief
Three platforms, three billing shapes, and the AI pricing change widened the gap between them. Quick directional read; the worked head-to-head lives in our Make vs n8n vs Zapier comparison.
Zapier wins on connector breadth and the lowest learning curve, and it is the on-ramp tool. AI steps cost more now, so it is the weakest of the three for high-volume AI reasoning loops.
Make trades a little breadth for a visual canvas and cheaper AI at scale. Make's Core plan runs 9 USD a month, and its AI Agents call MCP tools in one loop, which makes it the better-value pick for agentic work without standing up a server.
n8n deletes the per-execution meter entirely if you self-host, since n8n self-hosted carries no per-execution fee. For a Docker-capable team running AI loops at volume, that is the cost answer Zapier's task model can't match, at the price of running and patching the box yourself.
The verdict, by who you are
Zapier earns a 4.1 in June 2026, down from 4.3. The drop is mechanical: the June 15 AI model-tier pricing is a genuine cost increase on agentic steps with no offsetting feature gain, and Functions is being retired. The product itself did not get worse. The bill for one common use case did.
The picks, by situation:
- First-time, non-technical automation team. Zapier is still the pick. Connector breadth and the linear builder get you to a working automation faster than anything else, and if your Zaps are SaaS-to-SaaS plumbing, the AI pricing change never touches you.
- AI steps with tool calls as your primary use case, at volume. Re-evaluate. Run the formula on your real workflow, and if the multiplier hurts, switch the model tier to bring-your-own-key first, then compare against Make's 9 USD Core plan or self-hosted n8n before you renew.
- Team that needs the widest connector graph and a managed cloud. Zapier, on Professional or Team. You pay a premium on AI runs, but no competitor matches the integration count or the zero-ops reliability.
- Code-heavy logic builder. Look past Zapier. With Functions retiring, Code by Zapier covers steps inside a Zap, but Make's Code App or n8n's Code node treat code as a first-class building block.
If your situation is the first or third bucket, Zapier's plans start at 19.99 USD a month for 750 tasks, and you can start on Free to wire up your first Zap before paying. Just price the AI steps before you build the workflow around them.
FAQ
Did Zapier's AI step pricing really change? Yes. As of June 15, 2026, AI by Zapier steps are priced by model tier rather than a flat one task. Standard is 1x, Advanced (the default for steps with tools) is 3x, and Premium is 5x, applied to both the base run and each tool call (Zapier help, June 15 2026).
How many tasks does an AI step cost now?
Use the formula: tasks per run = (1 x model rate) + (tool calls x model rate). A default Advanced step with two tool calls is (1 x 3) + (2 x 3) = 9 tasks per run (Zapier help, model tier pricing).
How do I keep AI step costs down? Switch the model tier to bring-your-own-key, which keeps the multiplier at 1x while still allowing tools, or use Standard for prompts that need no tools. A three-tool agent costs 4 tasks on BYOK versus 12 on Advanced (Zapier help, model tier pricing).
Is Zapier Functions gone? It is being deprecated. No new signups since June 1, 2026, existing functions run until September 1, 2026, then stop. Migrate to Code by Zapier (Zapier help, Functions deprecation).
What does Zapier cost in 2026? Free is 0 USD for 100 tasks a month. Professional is 19.99 USD monthly (14.99 USD billed annually) for 750 tasks. Team is 69 USD monthly (52 USD billed annually) for 2,000 tasks. Enterprise is custom (Zapier pricing, pulled June 28 2026).
Ready to try it?
Try zapier →