Disclosure: some links on this page are affiliate links. If you sign up through one, we may earn a commission at no extra cost to you. We are not in Zapier’s or n8n’s affiliate programme, so nothing on this page earns us anything when you click through to either of them.
All five of these platforms advertise an AI feature on the cheapest account you can open — a free plan on four of them, and on n8n a trial that needs no card. We wanted to know something narrower than what the feature does: whether you can press run on a free account and get an answer back.
So on October 25, 2026 we opened the five accounts we hold. Four of them had something to press: three answered, and the fourth refused — and that fourth names the node we could not run on its own free plan feature list. On n8n there was nothing to press, because the trial that stands in for a free plan had already lapsed on our account, so n8n here is vendor documentation rather than a run. This page is what the screens said.
Quick Answer
Make was the only one of the five whose own built-in AI feature ran inside a workflow on a free plan. We placed the AI Agent node in a scenario, pressed Run once, and the module came back with the summary we had asked for. Its own result header read 1 operation – 1.05 credits – 1.5 KB, charged against the free plan’s monthly credit balance, which went from 719 to 711.95 in front of us.
Zapier and Pabbly answered too, but neither answer came from a workflow you could leave running. On Zapier the AI step ran as a preview inside an unpublished Zap: a real reply, and not one task charged for it on any of the three counters we checked afterwards. Publishing that Zap needs a paid plan. On Pabbly the builder in AgenticAI replied to our request and charged 7.18 AI credits for it, out of a wallet of 50 that the screen labels LIFETIME, and it created no workflow steps before stopping to ask us to connect Gmail.
Latenode refused, and n8n has no free plan to refuse with. Latenode’s free plan feature list names an AI Agent node; the node placed and configured without complaint, then the run failed with plug and play balance is empty before the model we had chosen mattered. n8n Cloud sells a trial rather than a free tier, and the workspace we hold has been paused since ours ended.
The Five Free Plans at a Glance
| Tool | What happened when we pressed run | Status |
|---|---|---|
| Make | The AI Agent node ran in a scenario and returned the summary. The module printed 1 operation and 1.05 credits as soon as it finished | Hands-on tested |
| Zapier | The AI by Zapier step returned a real answer in preview, in an unpublished Zap, and no counter moved. Publishing needs Professional or above | Hands-on tested |
| Pabbly Connect | The AgenticAI builder answered and charged 7.18 AI credits. It created no steps, and stopped to ask for a Gmail connection | Hands-on tested |
| Latenode | The run failed with plug and play balance is empty, and a modal offered only Manage Plan. The error row still logged 1 CPU second | Hands-on tested |
| n8n | Nothing to run. n8n Cloud has no permanently free plan; its trial needs no card, but ours has ended and our workflows sit paused | Researched |
| All five | Whether the three charges above can be read as a price comparison | Editorial judgment — no |
What We Tested
One AI request on each tool, on October 25, 2026, from accounts we already held. No payment method was added to any of them and nothing was upgraded, so what follows is what a free account could reach on that date.
Make, Zapier and Latenode were given the same fixed text, so that those three attempts differ by platform rather than by prompt:
Write a one-sentence summary of this message. Message: Hi, I filled in the form yesterday about the weekend workshop but I never got a confirmation email, and I want to check whether my place is still held.
Pabbly was asked something different, on purpose. Its AI feature is a builder rather than a step inside a workflow, so it received the same build request we had given Zapier’s Copilot on September 5 and the Pabbly builder itself on October 6: when a new row is added to Google Sheets, send an email with Gmail using the Email column as the recipient, the Name column in the subject and the Message column in the body. That makes Pabbly’s figure comparable with its own earlier one, not with Make’s.
Each tool was given its own built-in AI feature to run, on the vendor’s own model. Several of them also accept a model connection of your own — Latenode’s AI Agent node has a switch for it, and Make and n8n both document the same arrangement — and that is a different route with a different bill, which we did not take.
What this page does not cover: the quality of the answers, anything about paid plans beyond what the vendors print, and what a published AI step costs over a month. We ran one request each. That is enough to say what a free account reaches, and not enough to say what a month of it costs.
Make: The One That Ran Inside a Scenario
Make’s help pages say the AI Agents app is available on all plans using Make’s AI Provider, with the option to use custom AI provider connections on paid plans, and the free account matched that. The AI Agent node dropped onto the canvas with Make AI Provider selected by default. The tier list offered Small and Medium, marked Large as Paid, and marked every OpenAI model below them Paid as well.
We ran it on Small, the cheapest tier a free account can select, pressed Run once, and the panel reported Agent was thinking 1s before returning a one-sentence summary of our message. The module’s own header then printed the charge: 1 operation – 1.05 credits – 1.5 KB.
That figure is checkable in two further places on the same account. The organisation’s Credit Usage log holds a row reading Integration Make AI Agent, Type Scenario, Credits 1.05, timestamped to the second. The dashboard balance moved from 719 to 711.95 of 1,000, with six runs of an unrelated fifteen-minute scenario in between, each logged at exactly 1 credit. The organisation’s Credit usage row named what had run — Integration Make AI Agent, Type Scenario, Credits 1.05, recorded 2026/10/25 22:46:05 — and it landed in the Scenarios bucket, with the Agents bucket still at zero. So this was logged as the agent module inside a scenario, on Make’s own AI Provider. Both surfaces carry two decimal places, which is why a single run was enough to read. What the same free plan charges for a run with no AI in it is in our Make review.
Make publishes the rule as 1 credit per operation + credits based on AI tokens. Our run was one operation and 1.05 credits, so the charge carries both halves of that rule. We are not working out what the token half bought. Make does publish what a credit buys at each tier — 18,080 input tokens or 2,260 output tokens on Small and Medium — but the node shows no token count for the run itself, so any figure we put on it would be ours rather than theirs.
Maia, the builder that writes a scenario for you, is a different matter. Make puts it on all paid plans, and on free plans gives it a 30-day trial starting from signup. Ours had long since run out, so opening it returned Your free trial has ended. To continue building with Maia, upgrade to a paid plan anytime. The node and the builder are two different products with two different plan rules, and on an account past its first month only the node is usable. Where the other three builders put their own free-plan limit, and how far from the plan comparison each one keeps it, is in four AI builders and their free-plan gates.
The plan itself is a thousand credits a month with up to two active scenarios, read from the dashboard of Make’s free plan on the day. The AI Agents app is in open beta and Make says functionality and pricing may change, so treat the 1.05 as a reading taken on October 25, 2026 rather than a rate card.
Zapier: An Answer That Moved No Counter
Zapier’s pricing table lists AI by Zapier as unavailable on Free, and that is accurate about published Zaps. What it does not tell you is that the step still runs in the editor. The help page spells out the exception: on a free plan you can add AI by Zapier to an unpublished Zap to preview the results, where the tier defaults to Standard and the Advanced and Premium tiers are not available.
That is what we did. The prompt box carried a lightning badge reading 1 task, the preview returned a genuine one-sentence summary, and the panel beside it was greyed out with Tools require Advanced or Premium tier. The editor header read This Zap is using 1 Pro feature throughout.
Then we went looking for the charge, because a badge reading 1 task next to a free plan with a hundred of them is worth checking. The sidebar still read Tasks 0 / 100. Zap runs listed nothing from that day, the newest entry being October 1. The Task usage tab reported four billable tasks over thirty days, all of it to the left of the billing reset marker. We re-read the sidebar two minutes later in case the counter lagged. Three surfaces, none of them moved.
So a reader on Zapier’s free plan can watch an AI step work without spending anything. What they cannot do is ship it. Publishing a Zap that contains the step needs Professional or above, and on Free the step has no tools available to call, which means the published formula — Tasks used per run = (1 × model rate) + (number of tool calls × model rate) — collapses to a single task with nothing to multiply. The task itself is the unit we measured in the Zapier review.
One detail for anyone signing up fresh to repeat this: a new Zapier account does not start on Free. It starts on a fourteen-day Pro trial with no card required and drops to Free when that ends, so your first fortnight will show you more than a free plan does.
Pabbly Connect: The Builder Answered, Then Asked for Gmail
Pabbly’s AI builder has moved since we last used it. In the Connect dashboard, Create Workflow now offers Build with AI badged Opens Agentic AI, and clicking it opens a separate product at a separate address, with counters of its own. Anything written about the old route — Create Workflow, then Create using AI, inside Connect — describes a path that no longer exists under that name.
Inside AgenticAI the request was answered. The agent searched Connect’s app list twice, checked a webhook, listed our connections twice, and then stopped at an authorisation prompt: Authorize via Pabbly Connect – your credentials stay in Connect and never touch this chat, offering Connect Gmail or Not now. We took neither. The right-hand panel still read No steps yet.
The charge appeared in the chat header and itemised on hover: WORKFLOW AI CREDIT USAGE. Pabbly Provider only – Lifetime spend on this workflow. Source: Chat, Master Agent conversations. Credits 7.18. Requests 1. Total 7.18 AI credits. The wallet behind it moved from 42.98 available to 35.80, with lifetime used going from 7.02 to 14.20 against a lifetime purchased figure of 50.00 that did not change.
Two things there matter more than the number. The first is the word the screen uses for the total: LIFETIME PURCHASED, labelled all-time total. The fifty on our own free account are not a monthly allowance. Two builder requests have now taken 14.20 of them, and nothing we could find on the page says they come back.
The second is that the same prompt cost 7.02 on October 6 and 7.18 on October 25, with the same provider and the same stopping point. A Pabbly builder request is not charged at a flat rate. We are not going to tell you what moves it — we did not vary the prompt deliberately, and Pabbly shows no token counts — but two readings are enough to say it is not fixed.
Worth knowing about the free plan at Pabbly Connect: AgenticAI also meters runs in Units, 100 a month, and those do reset. Our request consumed none of them. Pabbly states plainly what a Unit buys — 1 unit covers up to 30 seconds of a run — and says nothing equivalent about what one AI credit buys. Two meters, two refill rules, one definition between them. The ordinary task meter those two sit beside is covered in our Pabbly Connect review.
Latenode: The Free Plan Lists the Node and Would Not Run It
Latenode’s free plan feature list, inside the product, names three AI items: an AI Agent node, AI Data Storage (RAG), and AI Code Copilot capped at ten requests. The AI Agent node carries no trial marker and no request cap. Reading that list, we expected this to be the easiest of the five.
The node itself encouraged that reading. Its model field holds 185 entries, and filtering for free returns six of them, marked with a green diamond where the paid models carry gold coins. Hovering a paid model ends its description with a pricing line — one read Pricing: 0.18 $/M input + 0.6 $/M output + 0.024 $/M cache read — and the six free entries have no pricing line at all. One model exists in both forms, paid and free, which is about as clear a signal as a product can give that a free lane exists.
We chose one of the six, sent the same summary prompt as Make and Zapier, and pressed Run once. The trigger completed. The AI node failed: plug and play balance is empty. A modal followed — PnP Tokens limit reached. You’ve either run out of PnP Tokens or reached your usage limit. Update your plan to continue. — with one button on it, Manage Plan, which we did not press.
The node is gated on a token balance before the model is considered, and a free Latenode account has no such balance. Not zero in the sense of spent: absent. The Statistics page reports Number of tokens in the current period: 0, but that is a consumption counter rather than a balance, and neither Balance block we opened lists tokens at all — only CPU seconds. The pay-as-you-go table prices a Plug and Play token at a dollar and shows no free row for it, where CPU Seconds shows 0–10,000 as free. Getting a token balance means moving to pay-as-you-go, and that screen asks for a card: To activate the plan, we’ll add at least $2.00 from your card to your wallet.
So the free plan line item is accurate in a narrow sense. You can place the AI Agent node, configure it, and pick one of its built-in models. You cannot run it on one. The free markers in that model list are not meaningless either — what they govern is the per-token price once you hold tokens, which we never got to test. The node also has a switch for a custom LLM connection of your own, and Latenode does not print what that costs or which plans may use it; it is a different route from the one we took.
The refused run was not quite free. Its history row reads Error, Dur. <1s, CPU-sec 1, Plug&Play 0, Oper. 2, and both CPU-second counters moved by exactly one afterwards, so the run was charged against the ten thousand the plan includes. That matches the documented minimum of one CPU second per run, and not the thirty that the in-product plan list prints — a disagreement inside Latenode’s own material that we have recorded rather than resolved. Latenode’s free plan is generous about runtime. It is the AI part that is not free. What a plain two-step run costs on that same CPU-second meter is in our Latenode review.
n8n: There Is No Free Plan to Test
n8n Cloud is the one tool here with no permanently free tier. Its pricing page lists a trial and paid plans, and the trial is printed as 1,000 executions, five concurrent executions, a 180-second timeout and one team project, with no credit card required. The page puts no length in days on that trial; the only fourteen days printed on it belong to the Business plan’s own trial. The console told us ours directly when we were inside it, counting down the days left. The execution quota that pays for running workflows, rather than for talking to the assistant, is the meter we looked at in the n8n review.
The trial carries an AI allowance of its own, printed as 800 Assistant credits, plus 100 AI credits. Assistant credits are defined in the documentation, and the definition is worth reading before you plan around them — they pay for conversations with the assistant and are never consumed by a workflow run, even one the assistant wrote. The model calls a workflow makes while it runs draw on a different pool, Gateway credits, and on nothing at all when the node uses your own credentials. The second figure on that line is harder to place: AI credits appear once on the pricing page, the paid plans list Assistant credits without them, and the documentation defines Assistant credits and Gateway credits but never an AI credit. In the documentation the phrase turns up in one place only — the message the editor shows when Assistant credits run out. We have been through how all five of these platforms meter their AI features separately.
We could not read any of this from a running instance. Our own trial has expired. The workflows were not deleted — they sit paused with their data intact, which is worth knowing if your trial is about to lapse — and the Cloud console reports that Assistant credits are not available for this workspace yet and that the feature is still rolling out. For self-hosted n8n the question changes completely, because the model calls then run on your own keys. That is a different setup and a different bill.
Two Feature Lists That Point the Wrong Way
Put Latenode and Zapier side by side and a pattern appears, and it is not the one we expected to find.
Latenode’s free plan advertises an AI Agent node and gives you a node that will not run on any of its built-in models. Zapier’s pricing table says AI by Zapier is not on Free and gives you a working preview that charges nothing. Both companies are describing the same kind of boundary — what a free account may do with an AI step — and both descriptions send a reader to the wrong conclusion, in opposite directions.
Our take, and this is a judgment rather than a measurement: a feature list is a statement about what exists in the interface, not about what will execute. The only reliable way we found to tell the difference was to press run on each one. Three of the five then printed their charge, or the absence of one, without being asked. That is a more useful thing to know about a platform than the length of its feature list.
Why These Three Numbers Are Not in One Column
We measured 1.05 credits on Make, 7.18 AI credits on Pabbly and no tasks at all on Zapier. It would be easy to sort those and call it a price comparison. It would also be wrong, for three separate reasons.
A Make credit, a Pabbly AI credit and a Zapier task measure different work, and no company publishes a conversion across them. The requests were not the same request: Make and Zapier summarised a message, while Pabbly was asked to build a workflow. And the three charges came out of three different kinds of pot — a monthly credit allowance, a lifetime wallet, and a monthly task allowance that did not move at all. The same five tools running one ordinary automation with no AI in it are in Google Sheets to Gmail, five ways.
What the three figures do support is narrower and still worth having. Each of these platforms will tell you what one AI request cost, in its own unit, on a free account, the moment the request finishes. Several will even price that unit for you — Latenode puts a Plug and Play token at a dollar, and Pabbly sells AI credits in packs with the price on the screen. What none of them publishes is the rate between one company’s unit and another’s, and that is the number a cross-tool comparison would need.
Which One Should You Use?
If you want an AI step running inside a workflow without paying, Make’s own node is the only one of the five that did it for us. It runs on any plan with Make’s own provider, the free plan carries a thousand credits a month, and the charge appears to two decimal places the moment the run ends. The AI Agents app is in open beta, so check the figure on your own account rather than trusting ours to hold.
If you only want to see what an AI step does before deciding, Zapier’s free preview was the only AI step we ran that moved no counter at all. It returned a real answer, and none of the three counters we checked afterwards had changed. Budget for Professional before you build anything that has to stay switched on, because the preview is where the free plan stops.
If you are weighing Pabbly, count the AI credits as a one-off endowment rather than an allowance. Fifty arrive labelled lifetime, two builder requests took 14.20 of them, and the per-request charge moved between our two attempts. The Units that meter workflow run time do reset monthly, and they are the counter with a published definition.
If you are going to Latenode for the AI node specifically, plan on pay-as-you-go from the start. The free plan’s runtime allowance is real and large, and the node’s built-in models are not reachable without a token balance, which needs a card. The scenario around the node still ran on the free plan and was charged a CPU second; what it would not do was the model call.
And if you are looking at n8n, the decision is about the trial rather than a free plan. No card is needed to start one, the executions are capped at a thousand, and when it lapses your workflows stay paused rather than disappearing. If you intend to self-host and bring your own model keys, none of what we measured here applies to you. We built the same two-step automation in both tools, on trials rather than free accounts, in our Zapier vs n8n comparison.
About these links: we are affiliates for Make, Pabbly and Latenode, and not for Zapier or n8n, which is why their links here are plain. The ordering above does not follow the money — the partner tool whose AI node would not run on its free plan is Latenode, and the one that cost a reader nothing at all is Zapier, where we earn nothing either way.
Everything on this page is an automation platform with AI inside it. The platforms where the product itself is the agent meter differently again, and we put five of those side by side in what AI agent platforms charge for.
FAQ
Can I run an AI step on a free automation plan at all?
On Make, yes — we ran one inside a scenario on the free plan and it charged 1.05 credits out of the monthly thousand. On Zapier you can preview one in an unpublished Zap without being charged, but not publish it. On Pabbly you can use the AI builder against a lifetime credit wallet. On Latenode the node will not run on its built-in models without a token balance, and n8n Cloud has no free plan.
Why did Latenode refuse when its free plan lists an AI Agent node?
Because its built-in models bill Plug and Play tokens, and a free account carries no token balance. The error was plug and play balance is empty, and it arrived before the model mattered — we had selected one of the six the list marks as free. For those models the feature list describes a node you can place and configure, not one you can execute.
Does a refused run still cost anything?
On Latenode it did. The failed run’s history row logged one CPU second against the free plan’s ten thousand, with Plug and Play at zero because nothing ran. We did not test failed runs on the other four, so we cannot tell you how they treat an error.
Can I publish a Zap with an AI step on the free plan?
No. The preview runs in an unpublished Zap at the Standard tier, and Zapier lists AI by Zapier as available on Professional, Team and Enterprise. Tools are blocked below the Advanced tier as well, so the free preview cannot call anything either.
Do Pabbly’s free AI credits come back each month?
The screen gives no sign that they do. The total is labelled LIFETIME PURCHASED and all-time total, in contrast with the Units counter on the same page, which reads 100 remaining this month and states that it resets monthly. Our two builder requests have taken 14.20 of the 50.
Was it the same request on all five?
Not quite. Make, Zapier and Latenode received the same one-sentence summary prompt. Pabbly’s AI is a builder rather than a step, so it got the same build request we had used on it before, which makes its 7.18 credits comparable with its own 7.02 from October 6. n8n received nothing, because there was no free plan to run it on.
Is there any way to try n8n Cloud without paying?
Yes — a trial, which needs no credit card and is printed as 1,000 executions with 800 Assistant credits. The pricing page puts no day count on it; the fourteen days it does print belong to the Business plan trial. The product itself showed us the days remaining while the trial was live. When ours ended the workflows were paused rather than deleted, with their data kept.
Sources and Verification
Hands-on, October 25, 2026, in accounts we hold, with no payment method added to any of them: a Make AI Agent module run with Run once on the free plan, read in the module result panel, the organisation Credit Usage log and the Make dashboard, which showed the free subscription’s thousand monthly credits and two active scenarios; an AI by Zapier step previewed in an unpublished Zap, with the Zapier home sidebar reading Free Plan, Tasks 0 / 100 and Zaps Unlimited, plus Zap runs and the Task usage tab, all three checked afterwards; a Master Agent request in Pabbly AgenticAI, reached through Build with AI in the Connect dashboard, with the credit tooltip and the AgenticAI Units screen — where the thirty-second definition is printed — read before and after; and a Latenode AI Agent node run with Run once, which failed, with the error text, the modal, the execution history row, the Statistics and Balance blocks and the free plan feature list inside the product read afterwards. For n8n we reached the Cloud console only; our trial has expired and we did not open a running instance.
Vendor documentation, read October 25 and 26, 2026:
- Make AI Agents app — available on all plans using Make’s AI Provider, with the option to use custom AI provider connections on paid plans, and the open beta notice that functionality and pricing may change. This page marks itself as a previous version and points to Make AI Agent (New), which repeats the same availability and states no charge, so the charges quoted above come from the credit-usage pages instead.
- Credit usage for AI agents — the current per-operation and token components, for Make’s own provider and for custom ones: 1 credit per operation + credits based on AI tokens.
- Updated Make AI Provider token pricing — in force from August 25, 2026: 18,080 input or 2,260 output tokens per credit on Small and Medium, and 3,616 input or 452 output on Large.
- Introduction to Maia — Maia is available in public beta on all paid plans. On Free plans, she’s available for a 30-day trial starting from signup.
- AI by Zapier model tier pricing — the free-plan preview in an unpublished Zap with the tier fixed to Standard, the plan table listing the feature on Professional, Team and Enterprise, the 1×, 3× and 5× multipliers, and the formula Tasks used per run = (1 × model rate) + (number of tool calls × model rate).
- Zapier pricing — the free plan’s 100 tasks a month, two-step Zaps and fifteen-minute update interval.
- Get started with your free Zapier trial — when you create a new Zapier account, you’re automatically enrolled in the Pro 14-day trial, that no payment method is required, and that the account downgrades to Free at the end.
- Pabbly Connect pricing — Start for free and get 100 tasks – no credit card required, and a two-step limit per workflow. The same free card also carries both Unlimited Workflows and You can create up to 2 workflows with up to 2 steps, so we are not printing a workflow count for the free plan.
- Latenode pricing — the free plan’s 10,000 CPU seconds, the AI Agent node line item, and the Plug and Play token at 1 token = $1 with no free row. This public page marks AI Code Copilot as Trial access; the ten-request cap quoted above is printed on the plan list inside the product.
- Latenode: custom LLM connection — the Use Custom LLM Connection switch inside the AI Agent node. The page states no plan requirement and no billing behaviour for it.
- Latenode: what are credits — One PnP token costs $1.00, that PnP nodes bill tokens on top of CPU seconds, and the one-CPU-second minimum per run.
- n8n pricing — the trial’s 1,000 executions, five concurrent executions, 180-second timeout and one team project, that no credit card is required, the 800 Assistant credits, plus 100 AI credits, and the absence of a permanently free Cloud plan. The fourteen days printed on the page belong to the Business plan trial, not to this one.
- n8n Assistant credits — running a workflow doesn’t use Assistant credits, even if n8n Assistant built it, and the editor’s out-of-credit message, You’ve run out of AI credits.
- n8n Gateway credits — what they cover at run time, and that nodes using your own credentials are unaffected when the balance reaches zero.
Where a figure came from a screen in our own account, it is that account’s figure. Where it came from a vendor page, the link above is the page we read. We have kept the dates on both, because the AI features on these platforms changed twice in the three weeks we spent watching them.
Comments