n8n vs Zapier for AI agents: one workflow priced both ways

n8n vs Zapier for AI agents is mostly a question of how each one bills a run and who will look after the workflow once it’s live. We priced the same lead triage workflow on both.

In short:

n8n charges per workflow execution, however many steps the workflow has. Zapier charges per task: each successful action uses one, and an AI step uses 1, 3 or 5 depending on the model tier.
For our lead triage workflow, 1,000 runs a month fits n8n Cloud Starter at €24 a month. On Zapier Professional, with four billable steps per run, it costs $133.50. At 10,000 runs it’s €60 on n8n Pro against $433.50 on Zapier. These are monthly-billing prices from each pricing page on 12 November 2026, and n8n prices in euros.
The model is the cheap part at this size. Claude Haiku 5.5 comes to about $0.40 per 1,000 leads on our token assumptions.
Zapier Agents is billed separately, in activities. Agents Pro gives 1,500 a month for $50, which covers roughly 375 runs of this workflow. Zapier says it’s moving Agents into AI by Zapier, which uses ordinary task pricing, and hasn’t set a date.
Git version control and separate environments are on n8n’s Business plan, which is self-hosted only, and on Enterprise. Starter and Pro don’t include them.
Self-hosted n8n has no licence fee for internal business use. Its licence isn’t open source in the OSI sense.

The example workflow

We picked lead triage because it’s short, it touches three tools many small sales teams already use, and only one step needs a model:

A new email arrives in the sales inbox.
A model reads it and returns JSON: whether it’s a sales lead, the sender’s first name and company, and what they need in one sentence.
If it isn’t a lead, the run stops.
The workflow looks up the sender’s email domain in HubSpot to see whether the company is already there.
It creates the contact in HubSpot.
It posts an alert to a Slack channel with the summary.

n8n vs Zapier for AI agents: which to pick

Pick	When	Watch out for
Zapier	The people building it aren’t developers, runs are in the hundreds or low thousands a month, and you want SAML single sign-on and shared app connections (Team plan)	Cost grows with every step and every run, and an AI step on an advanced model counts as 3 tasks
n8n Cloud	Someone technical will own it, and the workflow has many steps or runs often	Git and environments need a Business licence (self-hosted) or Enterprise. Run history is kept 7 days on Starter and 30 on Pro
n8n self-hosted	The data has to stay on your own servers and you can run a server	You own updates, backups and uptime, and the licence only covers internal use
Custom code	Volume is high, the logic needs tests and code review, or the platform bill keeps climbing	You own hosting, monitoring and every change to the APIs you call

How much does it cost at 1,000 and 10,000 runs?

n8n counts one execution per run, however many steps the run takes. Zapier counts a task for each successful action. Its pricing FAQ says triggers don’t count, and its help article on task usage says filters, Formatter and other built-in steps use none, and neither do actions that fail. An AI by Zapier step uses 1 task on a standard model, 3 on an advanced model and 5 on a premium one, plus more if the step calls tools.

Our workflow on Zapier uses one AI step, the HubSpot search, the HubSpot create and the Slack message. That’s 4 tasks per run on a standard model, or 6 on an advanced one. We assumed every email is a real lead, which is the most expensive case on Zapier, because a non-lead stops after the AI step.

Prices per month, billed monthly, read from n8n’s pricing page and Zapier’s pricing page on 12 November 2026:

	1,000 runs a month	10,000 runs a month
n8n Cloud	Starter, 2,500 executions: €24	Pro, 10,000 executions: €60
n8n self-hosted (Community Edition)	No licence fee, plus your server	No licence fee, plus your server
Zapier Professional, standard model (4 tasks a run)	4,000 tasks, on the 5,000-task tier: $133.50	40,000 tasks, on the 50,000-task tier: $433.50
Zapier Professional, advanced model (6 tasks a run)	6,000 tasks, on the 10,000-task tier: $193.50	60,000 tasks, on the 100,000-task tier: $733.50
Zapier Team, standard model	5,000-task tier: $178.50	50,000-task tier: $598.50

Notes on the table:

Both offer annual billing. n8n’s page says it saves 17% and Zapier’s says 33%.
n8n prices in euros and Zapier in dollars. We haven’t converted them.
10,000 runs uses all of n8n Pro’s 10,000 executions, so any growth means a bigger tier.
On Zapier, the model’s cost is inside the task price, unless you bring your own model, which costs 1 task plus your provider’s bill. On n8n and in code, you pay your model provider directly.

For the model, we assumed 1,500 input tokens per email (the instructions plus a typical email) and 500 output tokens, which includes the model’s thinking. At Anthropic’s prices for Claude Haiku 5.5, $0.10 per million input tokens and $0.50 per million output tokens, that’s $0.0004 per email: about $0.40 per 1,000 runs and $4 per 10,000. The build log will replace these assumptions with measured token counts.

For the costs that sit around the platform bill, such as hosting and the time someone spends reviewing what the workflow did, see AI agent cost per month.

What about Zapier Agents?

Zapier Agents is a separate product with its own meter. Zapier’s help centre says each trigger, action, knowledge lookup and web search an agent makes counts as one activity, and that using activities doesn’t touch your Zap task count. On the Agents tab of the pricing page, Agents Free includes 400 activities a month, and Agents Pro includes 1,500 for $50 a month billed monthly, or $33.33 a month billed annually.

Built as an agent, our workflow uses about four activities per lead: the trigger, the HubSpot search, the HubSpot create and the Slack message. 1,500 activities covers roughly 375 leads a month, so 1,000 runs is already beyond Agents Pro, and above that Zapier asks you to contact sales.

The same help article says Zapier is moving Agents to AI by Zapier, and the migration guide says AI by Zapier uses the same task-based pricing as the rest of Zapier. Zapier hasn’t set a date to turn off Agents. Zapier lists Agents under add-ons on its pricing page. At the volumes in our table, a Zap with one AI step is the Zapier option with a published price.

Errors, retries and what you see when a run fails

In n8n, each workflow can have an error workflow that runs when an execution fails. It starts with an Error Trigger node, and one error workflow can serve many workflows, so you can send every failure to the same Slack channel. Failed runs show in the Executions list. The pricing page lists automatic retries on every Cloud plan, and you can look back over run history for 7 days on Starter, 30 days on Pro and without limit on Enterprise.

In Zapier, Autoreplay and custom error settings are on Professional and above, according to the plan comparison. A failed action doesn’t use a task.

Version control, environments and handover

On 12 November 2026, n8n’s pricing page listed Git version control and multiple environments on the Business plan and on Enterprise, and showed both as not included on Starter and Pro. Business is self-hosted only and costs €667 a month billed annually or €800 billed monthly, with 40,000 executions. n8n’s source control docs explain that each environment is a Git branch, and that owners and instance admins can push and pull, while project admins can only push.

Zapier lists version history on Professional and above. Its pricing page doesn’t list a Git connection.

The Python version lives in a Git repository from the start, with the same review and deploy steps as any other code.

Whichever you pick, a handover needs:

the account and billing in the client’s own name, moved over if the builder set them up
a list of every credential the workflow uses (HubSpot, Slack, the model provider) and who can rotate it
the error workflow or alert, and the person who gets it
for self-hosted n8n, the server, the backups and the update routine
for code, the repository, the deploy step and the tests

Where Make fits

We didn’t build this in Make. This section comes from Make’s pricing page.

Make bills in credits. Each module action uses credits, most use 1, and some AI features use more. Routers and error handler modules use none. The free plan includes 1,000 credits a month. When credits run out, scenarios stop, and incoming webhooks wait in a queue until you add more.

By those rules, our workflow uses about five credits a run, because reading the email counts too. That’s about 5,000 credits a month at 1,000 runs and 50,000 at 10,000. Make’s pricing page didn’t show plan prices when we checked on 12 November 2026, so we haven’t priced it here.

When to move to custom code

Move when the platform bill grows faster than the work, when the logic has more branches than anyone can follow on a canvas, or when you need tests that run before every change. Code is also an option if the data has to stay on your servers and you don’t want to pay for n8n’s Business licence.

This is the core of our Python version. To keep it short, it leaves out the retries, logging and alerts a production version needs:

import json, os
import anthropic, requests

client = anthropic.Anthropic()  # reads ANTHROPIC_API_KEY
HUBSPOT = "https://api.hubapi.com/crm/v3/objects"
AUTH = {"Authorization": f"Bearer {os.environ['HUBSPOT_TOKEN']}"}
SCHEMA = {
    "type": "object",
    "properties": {
        "is_lead": {"type": "boolean"},
        "first_name": {"type": "string"},
        "company": {"type": "string"},
        "need": {"type": "string"},
    },
    "required": ["is_lead", "first_name", "company", "need"],
    "additionalProperties": False,
}

def triage(sender: str, subject: str, body: str) -> None:
    reply = client.messages.create(
        model="claude-haiku-5-5",
        max_tokens=2000,
        system="Decide if this inbound email is a sales lead. Give the sender's first name, "
               "their company and what they need, in one sentence.",
        messages=[{"role": "user", "content": f"From: {sender}\nSubject: {subject}\n\n{body}"}],
        output_config={"effort": "low", "format": {"type": "json_schema", "schema": SCHEMA}},
    )
    lead = json.loads(next(b.text for b in reply.content if b.type == "text"))
    if not lead["is_lead"]:
        return

    domain = sender.rsplit("@", 1)[-1]
    search = requests.post(f"{HUBSPOT}/companies/search", headers=AUTH, timeout=30, json={
        "filterGroups": [{"filters": [{"propertyName": "domain", "operator": "EQ", "value": domain}]}],
        "limit": 1,
    })
    search.raise_for_status()
    known = search.json()["total"] > 0

    contact = requests.post(f"{HUBSPOT}/contacts", headers=AUTH, timeout=30, json={
        "properties": {"email": sender, "firstname": lead["first_name"], "company": lead["company"]},
    })
    if contact.status_code != 409:  # 409 means the contact already exists
        contact.raise_for_status()

    status = "existing company" if known else "new company"
    requests.post(os.environ["SLACK_WEBHOOK_URL"], timeout=30, json={
        "text": f"New lead from {lead['first_name']} at {lead['company']} ({status}): {lead['need']}",
    }).raise_for_status()

Whichever version you run, someone has to update it when the model provider retires the model it calls. We explain why in why AI automations break after launch.

FAQ: is n8n open source, and is self-hosted n8n free?

Is n8n open source? Its code is public on GitHub, but most of it is under the Sustainable Use License. That licence lets you use or modify n8n “only for your own internal business purposes or for non-commercial or personal use”, and lets you share it with others only free of charge for non-commercial purposes. Files with “.ee.” in the name need a separate n8n Enterprise licence. The Open Source Definition says an open source licence “must not restrict anyone from making use of the program in a specific field of endeavor”, so n8n’s licence doesn’t meet it.

Is self-hosted n8n free? The Community Edition has no licence fee when you use it inside your own business. You still pay for the server and for someone’s time to keep it updated and backed up. SSO, Git version control and environments need a paid Business or Enterprise licence key, and the pricing page FAQ says that key checks in with n8n’s licence server every day.

If you want this workflow built on your own tools, that’s what our agent builds cover.