n8n Pricing Explained (2026): Cloud vs Free Self-Hosting

The one rule that explains n8n pricing

n8n bills on executions, and an execution is a whole production run of a published workflow, trigger to finish, no matter how many nodes, branches, loops or AI steps it contains^[1]. That's the opposite of per-task tools, where every step adds to the meter. A twenty-node n8n workflow that runs once is one execution. Since August 2025, every plan includes unlimited workflows, users and steps, so on Cloud the number that matters is how many times your workflows run per month. On the self-hosted Community Edition, even that has no quota.

The two paths

Self-hosted (Community Edition): free. You run n8n on your own infrastructure (Docker Compose is the usual route) with no execution quota and no licence cost. Your spend is your server bill, and n8n publishes no figure for it; its Hetzner guide says the smallest CPX11 type is enough for most usage, with 4 GB of RAM and 2 vCPUs needed only if you add the n8n Assistant. The trade is operations: you own updates, backups and security, and n8n recommends self-hosting for expert users. Community also lacks SSO, projects, workflow and credential sharing, Git and environments.

n8n Cloud: managed, per execution. n8n hosts everything on Azure in the EU (Frankfurt); you pay for a plan sized to your monthly executions. This buys back the operations work, which is worth real money if you don't have the technical capacity to run a server well. There is no free Cloud plan, only the trial.

n8n Cloud plans, decoded

Annual billing takes roughly 17% off each monthly figure (by our arithmetic, $20 against $24 is 16.7%). These are the USD prices n8n's US pricing page shows; the page displays prices for your browsing location, so readers elsewhere see local currency. There is no Business plan on Cloud, and Cloud can't upgrade to it. Above Pro-1 the Cloud path is Pro-2 at 50,000 executions, whose price n8n doesn't publish, then extra executions on Pro-2 billed monthly at $2 per 1,000 up to 500,000 a month, then Enterprise. There is no tier between 10,000 and 50,000.

The self-hosted licences

On annual billing the self-hosted Business plan is $9,600 a year for 480,000 executions, n8n's own figure. Larger Business volumes, per month on monthly billing, list at $1,182 for 50,000, $2,071 for 100,000, $3,167 for 200,000 and $6,294 for 500,000. The Start-up plan is for companies with fewer than 20 employees and less than €5M in total funding; the discount lasts a year, or longer while you still qualify. The Business trial runs 14 days and asks for a payment method.

What self-hosting actually costs at scale

"Free" self-hosting has a real cost curve once you grow. It's honest to name it. For modest volumes, a single server running n8n plus a Postgres database is all you need; regular mode has no concurrency limit unless you set one. When one instance isn't enough, you move to queue mode, n8n's scaling setup that spreads executions across worker processes, and it's included in the free Community Edition^[2]. Queue mode needs a Redis instance and a Postgres database^[2]. Filesystem storage for binary data isn't supported there, so n8n advises database mode; external S3 or Azure storage needs a Business or Enterprise licence. Standing it up is a migration project rather than a checkbox. You're trading money for engineering time.

How executions add up, and what inflates them (Cloud)

On Cloud, forecast executions by counting production workflow runs, then watch what counts and what doesn't:

• Triggers. A Schedule Trigger firing every minute is 43,200 executions in a 30-day month by our arithmetic. A polling trigger counts only when it finds new data; a webhook counts every request that activates it.
• Retries. Retrying a failed execution creates a new execution, so a flaky integration you rerun by hand adds to the meter.
• What's free. Manual runs in the editor, sub-workflows (only the parent counts), error-workflow runs (excluded on Cloud since n8n 2.38, September 2026), empty polls and rejected webhook requests don't count.

At the limit, Cloud executions are not queued and not billed as overage: they fail immediately with an execution-limit error while your triggers keep firing^[3]. Counts reset on the 1st of each calendar month, and Trial, Starter and Pro-1 get one free courtesy reset per customer. None of this applies to the self-hosted Community Edition, which has no quota; self-hosted Business has 40,000 a month.

Worked examples

A lean team, managed. A dozen workflows running a couple of thousand times a month fit inside Starter's 2,500; a few thousand more points at Pro's 10,000. Real bill on Cloud: Starter at $20 a month billed annually ($24 a month billed monthly), or Pro at $50 a month billed annually ($60 a month billed monthly). Self-hosted on Community. It's your server bill.

A high-volume shop. Say 100,000 executions a month. On Cloud that means Pro-2 billed monthly plus extra executions, or Enterprise. Fifty thousand extra executions at $2 per 1,000 add $100 a month by our arithmetic, on top of a base price n8n doesn't publish. Self-hosted Business at 100,000 executions lists at $2,071 a month on monthly billing. Self-hosted Community has no execution quota at all; you scale the server with queue mode and pay that infrastructure bill yourself. At high volume Community avoids execution charges entirely, in exchange for owning the ops.

An AI-heavy workflow. An agent that calls a model many times per run adds a model cost on top of the n8n execution, whether that's Gateway credits on Cloud or your own provider's bill. Budget both meters when an AI Agent loop is involved.

12-month total cost, illustrative

Over a year the Cloud plans are easy to total. Starter costs $240 a year billed annually or $288 a year billed monthly, by our arithmetic. Pro comes to $600 a year billed annually or $720 a year billed monthly. On the self-hosted side, Business is $9,600 a year on annual billing (n8n's figure for 480,000 executions), against $11,520 a year billed monthly if you pay the monthly rate. The Start-up plan is $4,000 a year billed annually (n8n's own figure), or $4,800 a year billed monthly (our arithmetic).

The row that doesn't fit a spreadsheet is the Community Edition. It has no licence cost and no quota, but you pay and run the infrastructure, and n8n publishes no figure for that. The real comparison for a technical team is a paid licence against the engineering time to run Postgres, Redis and a couple of workers, plus the features Community lacks.

The other bill: AI models and Assistant credits

AI-heavy teams face more than one meter. It's worth stating plainly. On Cloud Starter and Pro, Gateway credits (since n8n 2.36) let nodes use models from OpenAI, Anthropic, Google Gemini and others without your own API keys; they're prepaid, and nodes using them stop working when the balance hits zero. You can still bring your own key on every plan, and then your provider bills you directly. Assistant credits are a third, separate allowance for the n8n Assistant only: Starter 1,600, Pro-1 4,000 and Pro-2 9,600 a month since 1 October 2026, at one cent per credit.

Which plan fits you

The crossover: when self-hosting wins

The decision comes down to two questions: do you have the technical capacity to run a server well, and how high is your volume? If you're technical and running meaningful volume, Community self-hosting avoids execution charges. It adds data ownership. If you're not technical, or your volume is modest and your time is better spent elsewhere, Cloud's convenience is worth the subscription. You can start on Cloud and move later: workflows export as JSON, but exports don't carry credential secrets or execution history, so you recreate credentials after the move.

No per-seat pricing, a quiet saving

One thing your finance team will appreciate: there's no per-user charge. Every n8n plan includes unlimited users. Cloud plans bill on executions, the Community Edition is free, and paid self-hosted licences also bill on executions, so adding teammates to build and manage workflows costs nothing extra on any path. For tools that charge per editor or per seat, a growing automation team is a growing bill; on n8n, headcount and cost are decoupled. Sharing still has limits: Starter has one shared project, and Community has no sharing at all.

Trials, discounts, and avoiding surprises

Start free either way: self-host the Community Edition, or take the 14-day Cloud trial with Pro features and 1,000 executions, no credit card. The public API is off during the trial, and it expires on its own, so there's nothing to cancel. Verified non-profits get a one-time 30% discount code on all Cloud plans. Monthly plans are refundable within 14 days of the billing date if you cancel the account, and annual plans are non-refundable^[4]. Most customers are billed in EUR; US billing addresses are billed in USD. Commit annually only once n8n is clearly part of your stack.

The bottom line

n8n is strong value for automation, but the value depends on choosing the right path. If you're technical, the self-hosted Community Edition gives you executions with no quota for the cost of your own infrastructure, plus full data ownership. If you're not, Cloud's per-execution model, from Starter at $20 a month billed annually ($24 a month billed monthly), is predictable, and many-step workflows count as one execution. Model your executions, weigh your capacity to run infrastructure, and you'll land on the right one. See the full analysis in our n8n review, or start with the tutorial.