A price comparison for small-business owners choosing an automation platform. Every number comes from the vendor's own pricing page, and the sliders below price your own workload on all three.
By Joe Mascaro · Published · Updated · Prices verified , USDIt depends on how long your workflows are. For a handful of short ones, Make is cheapest at $16 a month for 10,000 credits, and its free plan runs two. Once a workflow has several steps, n8n pulls ahead, because it bills one execution per run however many steps it has: $24 a month for 2,500 runs. Zapier costs the most per step, from $29.99 for 750 tasks, and it's worth it when you need an app only Zapier connects to.
We've built on all three. For something quick and simple we'll often use Make, and for anything bigger, including AI agents, we build on n8n.
Triggers are free on all three. Filters and formatting are free on Zapier too, so count only the steps that touch an outside app.
All three panels share one scale, which resets as you move the sliders. Prices are for monthly billing on the cheapest plan that covers the workload, and each step in a line is one of that platform's plan tiers. Count only the steps that touch an outside app, since triggers are free everywhere.
| Zapier | Make | n8n | |
|---|---|---|---|
| Unlimited users on every plan | |||
| Runs on your own server | |||
| Branching costs nothing | |||
| A loop over 500 rows bills once | Per item | Per item | |
| Single sign-on under $100 a month | $69, yearly | $800 | |
| Keeps running past your quota | Held, replayable | Hard stop | Runs fail |
| Free plan you can build on | Two steps | Two scenarios | Trial only |
| App integrations | 10,237 | 3,500 | 2,329 |
n8n is source-available, which isn't quite open source. You can self-host it free for your own business, but you can't resell access to it, so the instance should be in your name.
Make renamed operations to credits on 27 August 2025, one for one, so older comparisons that say operations still hold. Some AI modules now cost more than one credit.
For AI work we use n8n, because it runs agents on the same plan and the same meter as every other workflow. We add a piece only when a job needs it, and we checked the dedicated agent platforms before settling on this.
Runs every workflow, including the AI agents, and keeps a log of each run.
Writes the drafts and makes the judgment calls. You pay Anthropic directly for what you use.
Added when a workflow has to remember customers or search your documents.
Added for heavy jobs, like reading scanned forms or processing large files.
Every account is set up in your name. That page shows how the pieces connect.
| Dify | $59/mo | Its credits are priced per model, so the bill is hard to predict. |
|---|---|---|
| Gumloop | $37/mo | It adds 8% to your model bill, and 16% if you bring your own key. |
| Activepieces | $20/mo | Its agents product sits under the paid commercial license. |
| Lindy | $29.99/mo | It charges per seat, so the bill grows with your team. |
| Relevance AI | Not listed | It took its prices off its public pricing page this summer. |
Unlike the platforms above, these are code libraries. A developer writes the agent, and it runs on a server someone pays for. Most small businesses never need one. They're sorted by GitHub stars as of 10 October 2026, which roughly tracks how many developers use each.
Best forTeams of agents that each play a role, set up quickly.
Best forEngineers who want control over every step and branch.
Best forPython teams that want the agent and its control panel from one vendor.
Best forTeams already building on OpenAI who want a light setup.
Best forTypeScript and Next.js product teams.
Best forGoogle Cloud shops and teams working in several languages.
Best forPython teams that want typed, testable agents.
Best forMicrosoft and .NET shops. It replaces AutoGen and Semantic Kernel.
Best forAgents that work in files and code, the way Claude Code does.
When a job outgrows n8n, we write that piece with the Claude Agent SDK and run it on Modal. n8n still starts it and logs it, so nothing changes on your side.
The Claude Agent SDK runs under Anthropic's commercial terms, and Mastra keeps its enterprise features under a paid license. The other seven are MIT or Apache 2.0.
LangGraph Platform is now LangSmith Deployment, Google's Agent Engine is now Agent Runtime, and Microsoft's AutoGen is in maintenance mode.
The software is free to run on your own server for your own business, including commercial use. n8n Cloud, where n8n hosts it for you, starts at $24 a month. What its license doesn't allow is reselling access to n8n or putting your own brand on it.
Usually, and by a lot. Make's entry plan is $16 a month for 10,000 credits, and Zapier's is $29.99 for 750 tasks. Both charge for every step on every item, so a workflow that loops over a long list costs more on either one than it would on n8n.
Zapier charges for each successful action, and its entry plan includes 750 of them. What you get for it is the biggest app library, 10,237 apps, and the easiest editor of the three. Triggers, filters, formatting and delay steps are free, so the real bill is lower than many comparisons claim.
A Zapier task is one successful action. A Make credit is one module run, and some AI modules cost more than one. An n8n execution is one full run of a workflow, however many steps it has, so five steps cost 5 tasks, 5 credits or 1 execution.
Zapier holds the extra runs so you can replay them, or bills each task if you've turned that on, at 1.25 times the normal rate on annual plans and 2.5 times on monthly. Make stops your scenarios until you add credits. On n8n Cloud's Starter and Pro plans, new runs fail until the 1st of the month.
Yes. n8n has an AI Agent node that calls a model and can use tools and memory, and each agent run counts as one execution like any other workflow. You pay the model provider, such as Anthropic, directly for what the agent uses.
For most small businesses, n8n Cloud. A $12 to $24 server saves a few dollars a month, and then someone has to keep it updated and backed up. We only self-host when a client has to keep data on their own infrastructure, which is the case with patient records.
It depends on the team. CrewAI and LangGraph are the most used in Python, Mastra is the main TypeScript option, and the Claude Agent SDK suits agents that work with files and code. Most small businesses don't need one at all, because an agent built in n8n covers the common jobs.
You should. We open every account in your name, you pay the vendors directly, and we work in it as an added user. If we stop working together, you remove our login and everything keeps running.
For most small businesses it's n8n and Claude, in accounts you own, and nothing more until a job calls for it.
Show us where your time goes and we'll send back a prioritized action plan, yours either way.