Key Takeaways
- A new AI by Zapier step on a paid plan defaults to the Premium tier: 5 tasks per run, plus 5 more for every tool call.
- The same 1,000-run lead-triage Zap uses 7,000 tasks on Premium and 3,000 on Standard.
- n8n counts one execution per workflow run regardless of steps; Make charges 1 credit per module action and also charges trigger checks that find nothing.
- Fix order: pick the lowest tier that passes your test, cap tasks per run, then decide whether your own API key is usable from Hong Kong.
A lead-triage Zap that used 3 tasks per run in May can use 7 tasks today without anyone touching it, simply because someone rebuilt the AI step. On 1,000 leads a month, that moves you from the 5K task tier to the 10K tier.
This is a cost breakdown, not a platform war: one workflow, priced on Zapier, n8n and Make by each vendor's own billing rules, checked on official pages on 18 November 2026.
How many tasks does an AI by Zapier step use?
It depends on the model tier. Standard costs 1 task per run, Advanced 3 and Premium 5. Each tool call the AI makes costs the same multiplier again. Zapier's help centre states that a new AI by Zapier step on a paid plan defaults to Premium, so an untouched step bills 5 tasks.
The formula from Zapier's model tier pricing page is: tasks per run = (1 × model rate) + (tool calls × model rate).
| Model tier | Multiplier | Tools allowed? | No tools | With 2 tool calls |
|---|---|---|---|---|
| Standard | 1x | No | 1 task | Not possible |
| Advanced | 3x | Yes | 3 tasks | 9 tasks |
| Premium (default) | 5x | Yes | 5 tasks | 15 tasks |
| Bring Your Own Key | 1x | Advanced or Premium models only | 1 task | 3 tasks |
Tasks come from one shared pool across Zaps, AI steps, code, MCP and the SDK, so the jump never shows as a separate AI line. Past your plan limit, Zapier can switch you to pay-per-task billing at a higher rate, per its pricing page.
Why did my Zapier task usage jump after June 2026?
On 15 June 2026 Zapier moved AI steps to model-tier billing. Steps created before that date stay on legacy pricing at 1 task, but any step you rebuild or duplicate is a new step and lands on Premium. Adding tools or knowledge sources multiplies the cost again.
The usual pattern: someone duplicates a working Zap, or rebuilds the AI step to add a knowledge source. The new step sits on Premium. Nothing breaks, and the task meter runs five times faster.
One more date to watch: 1 December 2026
From 1 December 2026, AI by Zapier called through Zapier MCP or the SDK will be billed at its own tier price instead of the MCP rate. If you call it from an assistant via MCP, recheck usage in early December.
What do 1,000 AI runs cost on Zapier, n8n and Make?
For one four-step lead-triage workflow run 1,000 times a month, Zapier uses 3,000 to 16,000 tasks depending on tier and tools, n8n uses 1,000 executions, and Make uses about 4,000 credits plus AI token credits if you use Make's built-in AI provider.
Test workflow: a web-form lead arrives, an AI step classifies it and drafts a reply, Gmail saves a draft, and Slack alerts sales.
| Platform and setup | Per run | Per 1,000 runs | Where the cost hides |
|---|---|---|---|
| Zapier, AI step on Standard | 1 + 1 + 1 = 3 tasks | 3,000 tasks (5K tier) | Trigger is free; each action is 1 task |
| Zapier, AI step on Premium (default) | 5 + 1 + 1 = 7 tasks | 7,000 tasks (10K tier) | Default tier on new steps |
| Zapier, Premium AI step doing Gmail and CRM lookup as tools | 15 + 1 = 16 tasks | 16,000 tasks (20K tier) | Each tool call adds 5 |
| n8n Cloud | 1 execution | 1,000 executions | LLM tokens billed separately |
| Make, webhook trigger + AI Toolkit on your own AI connection | 4 credits | 4,000 credits | Polling triggers charge every check |
On price, Zapier Professional starts at US$19.99 a month billed yearly for 750 tasks, and n8n Starter is €20 a month billed annually for 2,500 executions with unlimited steps, according to Zapier and n8n. Make's own help centre confirms 1 credit per module action, 2 credits per second for the Code module, and free routers and filters. Make shows plan prices interactively, so check the live card before you budget.
The Make cost almost nobody counts
A Make trigger that polls every 15 minutes uses 1 credit per check even when it finds nothing. That is 96 checks a day, or about 2,880 credits a month, before a single lead arrives. Zapier does not charge tasks for polling, and n8n does not count a run as billable until a workflow executes.
How do I lower the task cost of a Zapier AI step?
Open each AI by Zapier step, set the model tier to the lowest one that passes a ten-sample test, remove tools the AI does not need, and set a task limit per run. Then compare a week of Zap history before and after. Most classify-and-draft steps run fine on Standard.
- Find the expensive steps: open Zap history, sort by tasks used, and note every Zap with an AI by Zapier step.
- Open the Zap in the editor, click the AI by Zapier step, and open the Model tier dropdown. It shows Standard (1 task), Advanced (3), Premium (5) and Bring Your Own Key (1).
- Switch to Standard and run ten real past inputs through the step. If the labels and drafts are still acceptable, publish. If not, try Advanced before Premium.
- Move fixed actions out of the AI's tool list. If the AI always creates a Gmail draft, make it a normal Gmail action (1 task) instead of a tool call (3 or 5 tasks).
- Set the step's task limit per run. The default is 75 and the maximum is 500; when a live run reaches it, Zapier pauses the step for your approval.
When your own API key helps, and when it does not
Bring Your Own Key bills at 1x even with Advanced or Premium models and tools, so a 2-tool Premium step drops from 15 tasks to 3. The catch for Hong Kong: OpenAI and Anthropic do not officially support Hong Kong, and Google's Gemini API region list does not include Hong Kong either. Only use your own key if your company already holds an API account in a supported region or an enterprise cloud contract; otherwise Standard is the compliant saving.
Should Hong Kong teams switch to n8n or Make to save money?
Only if your workflows are long or high-volume. After fixing the tier, our test Zap costs 3,000 tasks per 1,000 runs, which fits a mid tier. n8n wins when one run has many steps; Make suits visual builders who avoid polling. Migration time usually outweighs savings on small, short Zaps.
| Tool | Usable from Hong Kong? | Billing unit | AI model access without your own key |
|---|---|---|---|
| Zapier | Yes; pricing page offers HKD billing | Task per action; AI by tier | Yes, built-in models |
| n8n Cloud | Yes; data hosted in Frankfurt | Execution per full run | Yes on Cloud plans ('AI models without API keys') |
| Make | Yes | Credit per module action | Yes, Make's AI Provider (token-based) |
The honest rule of thumb from these numbers: a 4-step Zap is cheap on any platform once the tier is right. The gap widens with steps. A 15-step reporting workflow run 1,000 times is 1,000 executions on n8n but roughly 15,000 tasks or credits elsewhere. Teams that must keep data in Hong Kong can self-host the free n8n Community Edition, at the cost of running a server.
Last verified: 18 November 2026, against Zapier, n8n, Make and Google official pages linked above.
What should you check this week?
Open your top five Zaps by task use and check the model tier on every AI step. A Premium step that only classifies or drafts is likely your biggest saving this month. Choose a platform by cost per 1,000 runs, not the sticker price.
Empowering AI. Centered on you.
Rather not maintain these workflows yourself?
Tier settings, task limits and API keys all need someone watching them. If you would rather not build and maintain this yourself, UD can deploy and run AI employees by role, with setup, configuration and day-to-day upkeep handled for you.
Get a Free AI Employee Assessment
Frequently Asked Questions
Does Zapier charge tasks when a Zap checks for new data?
No. Zapier's pricing page says polling for new data never uses tasks; only successful actions do. Filters, Formatter, Paths and Delay steps are also free.
Do failed Zap steps count against my task limit?
No. According to Zapier, an action that does not complete successfully does not count against your task limit. Check Zap history to see which steps actually ran.
Is n8n Cloud allowed for personal data from Hong Kong customers?
n8n Cloud stores data in Frankfurt, Germany. Under the PDPO you remain responsible for cross-border transfers, so check your privacy notice, or self-host n8n if data must stay in Hong Kong.
What happens when I run out of Make credits?
Scenarios stop until credits are added. Webhooks queue up to your plan's queue size, and you can buy extra credits in bundles of 1,000 or 10,000, according to Make's pricing FAQ.