Tasks, credits, executions: a glossary of what automation tools actually charge you for

Disclosure first: we make BizOps Pilot, one of the tools in this glossary. We are not a neutral review site. What we can offer instead is that every definition below is quoted from or sourced to the vendor's own documentation, every price was read off a vendor page on 22 July 2026, and anything we could not verify is marked UNVERIFIED instead of guessed at.

Here is the problem this page exists to fix. Ask five automation vendors what one "run" of your workflow costs and you will get five answers denominated in five different units. Zapier bills a task. n8n bills an execution. Make bills a credit. Gumloop also bills a credit, and it means something completely different from Make's. Lindy bills an allowance it does not define. Activepieces bills an active flow and does not count runs at all.

These are not synonyms. They are not even the same kind of quantity. Which means every "Zapier costs $19.99 and n8n costs €20, so they are about the same" comparison you have ever read is arithmetic performed on incompatible units. Nobody is lying. The units just do not line up, and the vendors have very little incentive to make them line up.

The one comparison that explains everything else

Take a three step workflow. Trigger fires, three actions run. You run it 500 times in a month.

• On Zapier, that is 1,500 tasks. Zapier counts each successful action step, so three actions times 500 runs is 1,500.
• On n8n, that is 500 executions. n8n counts one execution per workflow run no matter how many steps are in it.

Same workflow. Same work. Same month. One vendor's meter reads 3x the other's, before anyone has looked at a price.

Now add two steps to the workflow, which is a Tuesday afternoon decision that has nothing to do with volume. Zapier's number goes to 2,500. n8n's number does not move. On a per task platform, the cost of your automation is a function of how you built it. On a per execution platform, it is a function of how often it runs. That is the whole reason headline prices in this category do not mean anything on their own.

That example is an illustration, not research. We picked 3 steps and 500 runs to make the arithmetic legible. Your numbers are different, which is exactly the point of the method at the bottom of this page.

The glossary

Zapier: the task

Definition: a task is any successful action that runs in Zapier. Only successful actions count toward your task usage. (zapier.com/pricing, confirmed live 22 July 2026)

What counts as a task:

• Every successful action step.
• Any successful step inside an error handler path.
• Previously successful steps re run on a replay.
• A search action set to "proceed if nothing found" counts 1 task.

Sub Zaps are a genuinely unsettled point in Zapier's own documentation. An older help article said each action step inside a sub Zap, plus the "Call a Sub-Zap" and "Return From a Sub-Zap" steps, counted as tasks. Zapier's current task usage rates page instead lists "Sub-Zaps" itself among the built in tools that cost 0 tasks. The two pages do not obviously agree, and we could not reconcile them from the outside. Treat sub Zap task math as unverified until you check it against a live account, not as the precise breakdown we once published here.

What does not count:

• All trigger steps.
• Filter and Paths steps.
• Action steps that error or halt.
• Steps skipped because of a filter or path condition, or because the Zap errored.
• The "Start a Sub-Zap" trigger.

That "does not count" list is worth reading twice, because a large share of the affiliate content in this category asserts the opposite. If a comparison post tells you Zapier bills you for failed runs or for filter steps, the post is wrong and Zapier's own help centre is the thing that proves it.

The special cases where a task is not one task:

• AI by Zapier has been priced by model tier since 15 June 2026: Standard 1x with no tool calls available on that tier, Advanced 3x, Premium 5x, with tool calls on Advanced and Premium billed at the same rate as the step. New AI steps default to Advanced, which is the 3x tier. You can "connect your own AI account" (bring your own model) and pay 1x per step plus 1x per tool call, the cheapest tier that still supports tool calls. If a single step reaches 75 tasks in one run, the Zap pauses and asks for approval. (Zapier's task usage rates, confirmed live 22 July 2026, and Zapier help centre, 22 July 2026)
• Zapier Lead Router: each successfully routed lead is 5 tasks.
• Code by Zapier: 1 task per run at standard runtime, more if the runtime is extended.

Prices for scale: Free is $0 for 100 tasks a month. Professional starts at $19.99/mo and includes 750 tasks at that entry point when billed annually, or $29.99/mo billed monthly for the same 750 tasks. Team starts at $69/mo billed annually ($103.50/mo billed monthly) and includes 2,000 tasks at that entry point, its own floor rather than a price at the 750 task tier. Above your tier, Zapier switches you to pay as you go unless you turn that off, so the automations keep running and the bill keeps moving. (zapier.com/pricing, confirmed live with both the monthly and annual toggle, 22 July 2026)

Say the fair thing about it: the task is the most precisely documented unit in this glossary. Zapier tells you exactly what counts, exactly what does not, and publishes worked examples. Nobody else in this list is that clear. The unit is expensive per unit of work, and it is honest about being what it is.

Zapier MCP: two tasks per tool call

This deserves its own entry because it is a unit inside a unit, and because it is the one most likely to surprise you.

Definition, in Zapier's words: each successful tool call through your Zapier MCP server consumes two tasks. It is a fixed rate. Failed calls do not count. There is no separate MCP billing: it draws from the same plan allowance as your Zaps. Zapier's own MCP page states it plainly: "Each MCP tool call uses two tasks from your plan's quota, the same task bucket your Zaps use." (zapier.com/mcp, confirmed live 22 July 2026; docs.zapier.com/mcp/usage, read 16 July 2026)

The thing that makes this bite is that one instruction to an AI assistant is not one tool call. Zapier's own documentation supplies the examples, and they are more pointed than anything we would write ourselves:

Source: docs.zapier.com/mcp/usage, read 16 July 2026.

Zapier's free plan includes 100 tasks a month. At 2 tasks a call that is 50 tool calls a month, and the third example above spends 22% of the month's entire allowance on one sentence. Zapier's docs also note there is no limit on the number of tool calls per session, which means the meter is the only thing between an agent loop and your allowance.

We are MCP native and we do not meter, so we have an obvious interest in you reading that table a certain way. Read it however you like. The numbers are Zapier's.

n8n: the execution

Definition: one execution is one run of your entire workflow, regardless of how many steps it contains. n8n's pricing page puts it as workflow executions with unlimited steps. (n8n.io/pricing, read 22 July 2026; n8n docs, 23 June 2026)

What does not count as an execution (per those docs, 23 June 2026):

• Sub workflow executions. Only the parent counts.
• Manual runs from the editor.
• Error workflow runs.
• Polls that return no data.
• Malformed webhook requests that fail before the workflow starts.

Prices for scale, billed annually: Starter €20/mo for 2,500 executions. Pro €50/mo for 10,000. Business €667/mo for 40,000, and Business is self hosted. (n8n.io/pricing, read 22 July 2026. n8n's own pricing page still defaults to annual billing and would not render its monthly toggle state on a plain fetch, so these remain the annual figures. Check the toggle yourself before you rely on a monthly number.)

Say the fair thing about it, and it is a big one: the execution is the most customer friendly meter in this glossary, and it is not close. Charging per whole workflow run rather than per step means the vendor does not profit when you build your automation properly. Every exclusion in that list above is a place n8n chose not to bill you. If you are going to be metered at all, this is the version of metering that punishes you least.

And n8n self hosted has no execution meter at all. The Community Edition is free, source available, and unlimited on workflows and steps, because you are running it on your own infrastructure and n8n is not billing you by the run. Worth being precise about one adjacent claim rather than overstating it: n8n's own in editor AI Assistant, a separate feature from the workflow's own AI Agent nodes, is metered by "AI credits" on n8n Cloud today, and n8n's docs say that on self hosted "you'll bring your own API key, so no credit limits apply" once that Assistant ships there, which as of this writing is still "coming soon," not live. The execution meter is gone on self hosted regardless. The AI Assistant's own BYO key promise is real but not yet shipped for that deployment. If you are comfortable running and patching a container, that is an enormous amount of capability for the price of a small VPS, which is cheaper than we are. We are not going to pretend otherwise on our own glossary page.

Make: the credit

Definition, in Make's words: "Each module action in your scenario, like adding a Google Sheet row or fetching Gmail account data, counts as one credit." (make.com/en/pricing, confirmed live 22 July 2026)

So Make's unit is granular like Zapier's, not whole workflow like n8n's. It counts modules, not runs.

The part that matters more than the definition: the unit changed in 2025. Make announced on 13 August 2025 that it was replacing "operations" with "credits", effective 27 August 2025. Make's own framing, quoted from the announcement:

• "We're transitioning from operations to credits as our new billing unit."
• "This update allows us to introduce different credit values for different actions, meaning your billing will better reflect the specific actions you're performing."
• "The conversion is automatic... Your current plan and price stay exactly the same."
• "Most actions continue to consume 1 credit per module run, where advanced features, especially AI, will be priced at a fixed rate or varying by model and token usage."
• "Your credit usage may increase if you use Make's AI Provider."

(Make's official community announcement, 13 August 2025. Terminology confirmed settled on help.make.com/credits, page updated 19 January 2026, read 22 July 2026.)

Read that sequence carefully, because it is the defining move of this market. The sticker price did not change. The unit changed. And the new unit is elastic in the vendor's favour on precisely the workloads that are growing. Building an AI agent with your own connection also requires a paid plan. Make's own docs say custom AI provider connections are "paid plans only" without naming which paid tier is the actual floor, so do not repeat "Pro plan or higher" as a settled fact; the free tier is excluded and that is as precise as Make's own page gets.

The practical consequence is that a Make credit no longer has a fixed value, so you cannot multiply modules by runs and get your bill. You can get a floor. A three step scenario is at least 3 credits per run before you count the trigger module, any iterators, and any AI action's multiple.

Prices for scale, at the 10,000 credit tier: Free $0 for 1,000 credits a month. At 10,000 credits, Core is $9/mo billed annually or $10.59/mo billed monthly, Pro is $16/mo annually or $18.82/mo monthly, and Teams is $29/mo annually or $34.12/mo monthly. (make.com/en/pricing, confirmed live with both the credit-tier and monthly and annual toggles, 22 July 2026.)

What an elastic unit looks like when it goes wrong: a Make community user reported consuming 19,000 credits in a single agent run using Make's AI provider, calling the usage "absolutely ridiculous" (Make community, 3 June 2026). The Core plan includes 10,000 credits a month. That is one run costing roughly two months of allowance. This is a third party user report, not a verified cost study, and a replying community member reasonably attributed it to a non deterministic agent burning tokens working out where to go. Treat it as a worst case rather than a typical one. It is still a worst case that a customer could not have forecast.

Say the fair thing about it: Make's canvas is the most powerful visual builder in this category, and the credit's granularity is a direct consequence of that power. Routers, iterators and aggregators are real modules doing real work. Counting them is defensible. What is harder to defend is changing what a unit is worth while announcing that the price stayed the same.

Gumloop: also the credit, and it means something else entirely

Definition: a Gumloop credit is calculated on two things, per their docs:

1. Tokens analyzed. Every input and output token the AI processes costs credits, and output tokens cost more. Their docs put output at "2-4x more expensive" than input.
2. Tool calls. Tools that reach external data sources have their own credit cost. Tools that stay inside Gumloop, like sending a Slack message or reading a Google Doc, are free.

(university.gumloop.com and docs.gumloop.com, read 22 July 2026, no publication date on either page.)

So a Gumloop credit is a token meter wearing a credit's clothes. It is not a module count like Make's and not an action count like Zapier's. Same word, different animal.

Gumloop's docs are unusually honest about the failure mode, and we should credit them for it rather than pretend we found it. Two quotes, both theirs:

"Agent credit costs are variable. The same agent might cost 2 credits for a quick question and 200 for a deep research task."

"There's a hidden cost to watch for: when a tool returns data, the model has to read the entire response to use it. A tool that returns a large JSON payload... can generate thousands of input tokens just from the result. Sometimes analyzing the response costs more than the tool call itself."

A unit whose own vendor documents a 100x spread for the same agent is not a unit you can budget with. That is not a criticism of Gumloop's honesty. It is a criticism of the model that requires the honesty.

Fixed rates they do publish: image generation is 30 credits per image, 15 with your own key. Audio transcription is roughly 1 to 2 credits per minute, halved with your own key.

The bring your own key term, which is the important one: "Provide your own provider API key to cut model costs. BYOK gives you 50% off AI model credits in both agents and workflow AI nodes." It requires the Pro plan or higher. (docs.gumloop.com, read 22 July 2026, undated.)

Read that literally. You supply the API key. You pay OpenAI or Anthropic directly for every token. And Gumloop still charges you half the credit meter on top. You pay the model provider, then you pay the meter. It is in their own documentation.

Prices for scale: Free $0 for 5,000 credits a month. Pro is $37/mo billed monthly, or $30/mo billed annually, for 20,000+ credits with unlimited seats. (gumloop.com/pricing, confirmed live with the monthly and annual toggle, 22 July 2026.)

What the spread looks like when it goes wrong: a Gumloop user reported an agent searching a GitHub repo via MCP cost 900 credits in one run, and that their first test "burned everything" on their original account (Gumloop forum, 12 January 2026). On the free tier's 5,000 credits that is about five runs. Again: one user's report, not a study.

Say the fair thing about it, and it costs us to say it: Gumloop Pro is $37/mo, which is cheaper than our $39, with unlimited seats, hosted MCP, and $70M of outside funding behind it (TechCrunch, 12 March 2026; SiliconANGLE, 13 March 2026). They are AI native in a way we are not. If you want AI first pipelines and the credit meter never bites you, they are a better product for you than we are, and they cost less.

Lindy: the allowance nobody defines

Definition: a credit, though not on the pricing page. Lindy's billing docs, not its pricing page, name the unit and give a per task cost range: "Most tasks cost 1-3 credits on basic models and ~10 credits on large models. Every task is at minimum 1 credit," and cost rises with "model intelligence," so a better model costs more credits, not fewer. (docs.lindy.ai/account-billing/credits, read 22 July 2026)

What the pricing page itself still gives you is only relative: Plus at $49.99/mo with "Standard" usage, Pro at $99.99/mo with "3x more than Plus", and Max at $199.99/mo with "7x more than Plus". (lindy.ai/pricing, page self reports last updated 9 May 2026, read 22 July 2026.)

Three times what? Seven times what? We now know the unit is a credit. We still do not know how many of them a dollar buys on any tier. Marked UNVERIFIED for the one number that would actually let you budget. Lindy does not appear to publish an absolute monthly credit allowance anywhere we could find, on the pricing page or in its billing docs.

A single 2025 review on a third party site referred to "$50/month for 5,000 credits", which would put Plus at 5,000 of something, but that is one undated user claim against a pricing page that has since been rewritten. We are not treating it as a fact and neither should you.

If you are evaluating Lindy seriously, ask them what the unit is before you sign anything. That is not a dig. It is the same advice we would give about any unit you cannot read off a page.

Say the fair thing about it: Lindy has also largely left this category. It now sells a per seat personal AI work assistant priced by inboxes rather than a workflow automation platform (Lindy Assistant launch, 10 February 2026). If what you want is an assistant that triages your inbox rather than a pipeline that runs unattended, it is aimed at you and we are not.

Activepieces: the active flow

Definition: Activepieces charges "$5 per active flow per month" with "Unlimited runs" stated explicitly, and the first 10 active flows are free. (activepieces.com/pricing, confirmed live 22 July 2026.)

An active flow is a flow that is switched on. That is the entire unit. It is a number you set on purpose, you always know what it is, and it does not move when a webhook misfires at 3am.

We are including this entry even though it is not a run meter, because it is the proof that a run meter is a choice rather than a law of physics. Activepieces prices the thing you build, not the number of times it runs.

Say the fair thing about it, and this one costs us too: Activepieces is the closest commercial rival to our own pricing model and it is arguably a cleaner expression of it. At ten flows or fewer it is free, and free beats $39. If you have a small number of very high volume automations, price it out against us properly. You may well find they are cheaper.

BizOps Pilot: no unit, which is the point and also the catch

We have no row in the table above, because we do not have a unit. We do not count runs, tasks, actions, steps, tokens, tool calls, or minutes. There is nothing to run out of.

The prices: free beta, Pro $39/mo, and a Founding option at $249 once for a lifetime seat, hard capped at 300 seats and enforced at checkout. AI inference runs on your own API key and you pay Anthropic or OpenAI directly, at their prices, with nothing added by us.

The free plan bounds how many Pilots and Systems run at once: three active Pilots, one active System. That is a capability ceiling, not a meter. An active Pilot on the free plan can run forever, as often as its trigger fires. Runs are unlimited on every plan, including the free one. The REST API and the MCP server are the Pro features.

Now the catch, stated plainly, because a glossary that lists everyone else's tradeoffs and not our own is an advert.

A meter is not just a way to charge more. It is a way to move variance onto the customer. When Make's user burned 19,000 credits in one run, Make's costs went up and so did that customer's bill. The meter transferred the surprise from the vendor's P&L to the customer's. That is the actual function of a credit.

We do not do that, which means we absorb that variance ourselves. The 19,000 credit run costs us money and costs you $39. So does the next one. That is the deal, and it is a real risk we are taking, not a clever trick:

• Bring your own key is what makes it survivable at all. The volatile, expensive part of an AI workflow is the tokens, and those are on your bill and your provider account, not ours. If we ever shipped a "we supply the AI key" convenience inside the flat plan, we would reintroduce a marginal cost that scales with your usage, and a meter would follow within the year. That is the one move that would invalidate the model. We are not going to make it.
• Our real costs still scale with heavy users. Compute, queue, database, storage, egress, and support all go up when you run more. A customer using us 100x more than another pays exactly the same $39. That works because we are small and our costs are near zero per run, and it is genuinely untested at the scale we would like to reach.
• Flat pricing selects for exactly the customers the metered vendors are pricing out. The person running the agent loop that burned two months of Make's allowance has a strong financial reason to come here. We know that. We built the pricing anyway.
• If we ever get this wrong, the honest answer is concurrency, not counting. Every vendor in this glossary already caps concurrency somewhere. That is the lever we would reach for, because it keeps "no meter to run out" literally true. We would rather tell you that now than have you discover our position on it later.

So the honest version of our pitch is not "meters are evil." It is: the variance is real, somebody eats it, and we have decided it is us. You are entitled to price in the risk that we are wrong about being able to afford that. We would rather you did that with the numbers in front of you.

The method: how to actually compare two vendors

The units cannot be converted into each other. So do not try. Do this instead.

1. Write down your three highest volume workflows. Not your favourite ones. Your loudest ones.

2. For each, count two numbers: steps per workflow, and runs per month. The step count is the number every calculator on the internet omits, which is why every calculator on the internet is wrong.

3. Multiply into each vendor's own unit. Do not compare their prices. Compare your workload, expressed in their unit:

4. Look up that number on the vendor's price slider, and only now compare dollars to dollars.

5. Then do it again for the version of your workflow you will have in six months, when it has two more steps and fires three times as often, because that is the bill that actually decides whether you stay.

Notice what happens when you run this properly. Half the vendors in this category cannot be priced by their own customer in advance. Zapier, n8n and Activepieces can be. Make can be given a floor. Gumloop and Lindy cannot be priced at all until you have run the thing for a month and looked. That is not a scandal. It is a fact you should know before you build on top of one of them.

And if you finish this exercise and the answer is n8n self hosted, or Activepieces at ten free flows, or Zapier because the catalogue has the one app you cannot live without, then that is the answer. We would rather lose the sale than win it with arithmetic we know is bent.

What we could not verify

In the spirit of the rest of this page.

Zapier's Team tier base task allowance is now resolved, not UNVERIFIED. As of 22 July 2026 the pricing page states it directly: 2,000 tasks/mo at $69/mo billed annually. We are removing the earlier UNVERIFIED flag on this one figure because we found the number, not because we stopped checking.

Sub Zap task counting on Zapier remains genuinely unsettled, and is now UNVERIFIED where our earlier draft stated a specific breakdown as fact. An older Zapier help article and the current task usage rates page do not obviously agree on whether action steps inside a sub Zap bill as tasks. See the task section above.

Lindy's absolute credit allowance per tier is still not published anywhere we could find, on the pricing page or in Lindy's own billing docs. We now know the unit name, a credit, and the per task cost range, 1 to 3 credits on basic models and about 10 on large models, but not how many credits "Standard", "3x" or "7x" usage actually equal. UNVERIFIED.

Monthly versus annual toggle state for n8n still was not resolved. n8n's pricing page defaults to annual billing and would not render a monthly figure on a plain fetch as of 22 July 2026. The n8n prices we quote remain the annual figures; check the toggle yourself before relying on a monthly one. Make's and Gumloop's monthly and annual prices are both now confirmed directly in this page, live, as of the same date.

Make's bundled credit count at the entry price is quoted at what we read on the pricing page's live credit tier and billing toggles on 22 July 2026, and nothing more. Third party sources have historically reported figures that are not arithmetically consistent with the live page or with each other; we did not rely on any of them.

The two agent burn reports (19,000 credits on Make, 900 on Gumloop) are user forum posts, not studies. They are dated and linked, they are consistent with each vendor's own documented variability, and they are still single anecdotes. We are using them as illustrations of a documented mechanism, not as typical costs.

Gumloop's and Lindy's docs carry no publication dates. We read them on 22 July 2026. That is the only date we can honestly attach.

Sources

Prices and definitions for Zapier, n8n, Make, Gumloop, Lindy and Activepieces were re-read and reconciled on 22 July 2026. Where a page carries its own publication or update date, that date is noted separately below.

Zapier

• Pricing, tiers, pay as you go behaviour, task definition and counting rules, confirmed live including the monthly and annual toggle: https://zapier.com/pricing
• Task usage rates, the current permanent rate table, confirmed live 22 July 2026: https://zapier.com/pricing/rates
• Professional at $19.99/mo annual ($29.99/mo monthly) including 750 tasks: https://zapier.com/pricing
• AI by Zapier model tier pricing, default tier, formula, 75 task circuit breaker: https://help.zapier.com/hc/en-us/articles/46425475442829-AI-by-Zapier-model-tier-pricing
• Zapier MCP, 2 tasks per successful tool call, confirmed live 22 July 2026: https://zapier.com/mcp
• Zapier MCP usage worked examples: https://docs.zapier.com/mcp/usage

n8n

• Pricing and execution definition: https://n8n.io/pricing/
• What counts as an execution, 23 June 2026: https://docs.n8n.io/build/understand-workflows/understand-executions
• Self hosted Community Edition guidance: https://docs.n8n.io/choose-how-to-use-n8n#self-hosted-community-edition

Make

• Pricing and credit definition, confirmed live with the credit tier and monthly and annual toggles: https://www.make.com/en/pricing
• Operations to credits announcement, 13 August 2025, live 27 August 2025: https://community.make.com/t/introducing-credits-a-new-system-of-billing/89480
• Credits terminology, page updated 19 January 2026: https://help.make.com/credits
• Custom AI provider connections restricted to paid plans, 1 credit per operation even on your own key: https://help.make.com/credit-usage-for-ai-agents
• 19,000 credit agent run, user report, 3 June 2026: https://community.make.com/t/need-help-with-agentic-workflow/110223

Gumloop

• Pricing, confirmed live with the monthly and annual toggle: https://www.gumloop.com/pricing
• Credits, "2 credits or 200" quote: https://docs.gumloop.com/core-concepts/credits
• Tokens and costs, output 2-4x input, internal tools free: https://university.gumloop.com/ai-fundamentals/tokens-and-costs
• BYOK at 50% off AI model credits, Pro plan required, image and transcription rates: https://docs.gumloop.com/core-concepts/ai_models
• 900 credit GitHub MCP run, user report, 12 January 2026: https://forum.gumloop.com/t/github-mcp-burning-hundreds-of-credits/3553
• $50M Series B led by Benchmark, 12 March 2026: https://techcrunch.com/2026/03/12/gumloop-lands-50m-from-benchmark-to-turn-every-employee-into-an-ai-agent-builder/
• $70M total outside funding, 13 March 2026: https://siliconangle.com/2026/03/13/gumloop-reels-50m-ai-automation-platform/

Lindy

• Pricing, page self dated 9 May 2026 as of this reading: https://www.lindy.ai/pricing
• Credits, unit definition and per task cost range: https://docs.lindy.ai/account-billing/credits
• Lindy Assistant repositioning, 10 February 2026: https://www.lindy.ai/blog/lindy-assistant-launch

Activepieces

• Pricing, "$5 per active flow per month", "Unlimited runs", 10 free flows, confirmed live 22 July 2026: https://www.activepieces.com/pricing

Pricing in this category changes without notice, and as this page documents, the unit changes without notice too, which is harder to spot. Every number here is a dated snapshot, not a promise. Re-check before you buy anything, including ours.