Make credits explained: what a scenario really costs in 2026

If you looked up Make.com credits pricing, the plan table is the second question. The first is how many credits a scenario spends when it runs. A quiet polling watch can cost more than the modules you remember adding.

This is a reading of Make’s help and pricing pages, checked on 29 October 2026. It does not quote plan prices or included credits. Those sit on Make’s pricing page, and they move. The sums below apply documented rules to hypothetical workloads. They are not figures from a live account. If you are still choosing a tool, start with Make vs Zapier vs n8n.

Credits are not operations

Introducing credits made credits the billing unit and converted old operations one for one. What a credit costs still depends on the plan. Credits are what you buy. Operations are module runs. On a normal app, one operation is one credit. Toggle credits beside the scenario name: the coin on each module is the figure to trust.

A bundle is one submission, email or file. A trigger runs once per check, however many bundles come back. Later modules run once per bundle.

What consumes a credit

The classes are from How features use credits.

A polling trigger, often named Watch, costs one credit per check, including a check that finds nothing. Make calls that empty run a check run. It shows in scenario History. Types of modules defines the two trigger styles. Step 10 uses the same trigger-only sums as this article: every five minutes is 288 checks a day, and 8,640 in a 30-day month. Hourly is 720. Once a day is 30. A 31-day month adds a day’s checks.

An instant trigger is a webhook, described on Webhooks. You pay when a request arrives. Look for the lightning icon and the Instant label. Without it, you are polling. Ten bundles from one trigger run are still one credit. Later modules are where ten gets expensive.

A search is one credit however many rows return. In Make’s sheet example, ten rows cost one credit, and the action that deletes them costs one credit per row, so 11 in total. An iterator is one credit to split an array, then one credit per item on each later module: five attachments are one plus five uploads. An aggregator is one credit per aggregation, not per row, and only inside that run. Make rewrites a watch plus ten emails (11 operations) as a watch, one aggregator and one email (3). Separate webhook runs do not collapse. For ordinary modules, those operation counts are the credit counts.

These cost nothing: routers, filters, the error handlers Rollback, Break, Resume, Commit and Ignore, plus Call a subscenario, Start a scenario, Return output, Customize run name, and Make Functions. The pricing FAQ names only routers and error handlers. A filter that stops a bundle keeps later modules from running. Modules inside a called scenario still cost. A free handler does not zero the module that failed. Read the coin. A retry is another run. Exponential backoff schedules some of them.

AI modules

Your own key on OpenAI, Claude or Gemini is one Make credit per operation. Tokens sit on the provider’s invoice, not the coin. Make documents that custom connection as a paid-plan feature.

Make’s AI Provider, in Make AI Toolkit and Make AI Agents, adds token credits on top of that operation. Since 25 August 2026, Small, Medium and Large use separate input and output rates. The token update explains it. The live table is on the credits page, and Make says the rates can change. Automatic connections, including Make AI Web Search, some Content Extractor modules, and Simple Text Prompt in the OpenAI, Gemini, Groq and Claude apps, can also bill on file size, pages or time. Simple Text Prompt is not one credit just because it sits in the OpenAI app.

Two hypothetical months

Both examples use a 30-day month and one credit per non-AI operation. Recompute them. Do not treat them as a quote.

Lead capture

The shape is lead capture from a form to a CRM: filter junk, classify the rest, match the email, create or update, notify you. A person still sends the reply.

Assumptions. 400 submissions in the month. The filter is immediately after the trigger and drops 80. The AI step uses your own connection, so the model invoice is outside this sum. The router is free and only one write runs. Each webhook delivery is its own run.

A genuine submission is trigger 1, filter 0, AI 1, search 1, write 1, notification 1: 5 credits. A dropped submission is the trigger only: 1 credit.

• 80 dropped × 1 = 80
• 320 genuine × 5 = 1,600
• Webhook month = 1,680 credits

Polling every 15 minutes, the interval in Make’s beginner lesson, changes the trigger only. Checks are 24 × 4 = 96 a day, and 2,880 in 30 days. Do not add 400 trigger credits on top of that. One check is one credit even when several submissions return together. Downstream stays 320 × 4 = 1,280. The polling month is 2,880 + 1,280 = 4,160 credits.

The gap of 2,480 is the empty checks: 2,880 − 400. Classifying before the filter would add 80 credits on either design. Put the free test first. An aggregator does not shrink the webhook version, because each submission is already its own run.

Email triage

This is the Make sum for the n8n email assistant: classify a slice of mail and leave a draft. The Shopify version adds a lookup. The mailbox module polls. 600 messages arrive. A filter lets 150 through and drops 450 before the model. The AI step is your own connection. One action creates the draft. The router is free.

Naive: every five minutes, no filter, a draft and a chat ping for every message.

• Trigger: 12 × 24 × 30 = 8,640
• AI, draft and ping on 600 messages = 1,800
• Total = 10,440 credits

Tighter: hourly, filter to 150, no ping.

• Trigger: 24 × 30 = 720
• AI 150 + draft 150 = 300
• Total = 1,020 credits

The difference is 9,420. Fewer checks are 7,920 of it (8,640 − 720). Not classifying or drafting the 450 filtered messages is 900. Removing the ping is 600. The schedule dominates.

The same 150 drafts on Make’s AI Provider

Keep the tighter scenario and swap the AI step to Make AI Toolkit on Make’s AI Provider, Medium tier. On 29 October 2026 the credits page listed that tier at 18,080 input tokens per credit and 2,260 output tokens per credit, plus one credit per operation. If the live table has moved, redo the division.

Token assumption, not a measurement. 800 input tokens and 120 output tokens per message: a short prompt, a short email and a short JSON reply, using Make’s own gloss that an English token is about four characters. This site did not run it.

• Operations: 150 × 1 = 150
• Input: 120,000 tokens ÷ 18,080 = 6.637 credits
• Output: 18,000 tokens ÷ 2,260 = 7.965 credits
• AI total = 164.602 credits

That is about 164.6 Make credits for the AI steps, against 150 plus a separate provider invoice if you bring your own key. Output tokens are the dearer column, so a long draft from a short prompt can cost more. The help pages checked for this article do not state a rounding rule, so 164.602 is the sum of the fractions. Read one real coin before you scale it. With the hourly trigger and the draft action, the month is about 720 + 164.6 + 150 = 1,034.6 credits.

How to spend fewer credits

Use an instant trigger when the app has one. Schedule a scenario sets “as data arrives”. Filter before the model: a honeypot, an empty body, mail you sent, or a newsletter label costs nothing, and what it stops is never classified. Aggregate only inside one run, and only if one summary will do. Lengthen a poll that can wait, and switch off a test scenario. Lower a search Limit so the action after it does not walk every row.

When the credits run out

Make’s pricing FAQ says scenarios stop until credits are added. Webhooks queue up to the plan’s allowance. The webhooks page says further calls are rejected once that queue is full. Afterwards, the queue is processed, and polling looks for records since the last successful run. Included credits expire at the end of the term and do not roll over.

Running out early means upgrading, buying extra credits, or auto-purchasing where the plan offers it. Extra credits says top-ups cost more than in-plan credits, that a failed payment pauses scenarios, and that exhausting data transfer can buy credits even while some remain. On the Core plan billed annually, that page says the allowance resets monthly and extra credits expire with it. Read the page for the units and the premium.

When Zapier or n8n fits better

Stay on Make when someone will read the canvas, the trimmed volume still fits a plan you have opened, and nobody wants a server.

Zapier fits a short line whose owner will not learn bundles. Its pricing page calls the unit a task. Read that definition. Five tasks on every email is the same shape as five Make modules: narrow the trigger and filter early.

n8n Cloud fits when one item walks through many modules. n8n’s pricing page says one workflow execution is one run of the whole workflow, however many steps it has. Confirm that wording. It does not remove empty checks. n8n’s own note treats an every-five-minutes schedule as thousands of executions a month.

Self-hosting moves the work off Make credits and onto a server, backups and your time. Paid self-hosted plans can still meter executions. Follow the Docker guide only if someone will patch the box.