Some links in this article are affiliate links. Automated Wealth Lab may earn a commission at no extra cost to you. See our Affiliate Disclosure.
Disclosure: This article contains affiliate links to n8n, Make and Relevance AI. If you sign up through them, Automated Wealth Lab earns a commission at no extra cost to you. It never changes what we recommend — the affiliates only exist where the tools genuinely fit the workflow.
Agent builders don’t sell you agents. They sell you meters — and each platform’s meter counts a different thing. n8n counts a whole workflow run as one unit no matter how much work happens inside it. Make counts every module that fires. Relevance AI counts every tool run and runs a second meter for the AI itself. Comparing their plan prices ($20, $9, $19 a month) tells you almost nothing, because the same finished job consumes wildly different amounts of each meter.
So this benchmark does the only comparison that matters: it defines one completed task, precisely, then prices that task on all three platforms at two volumes — platform cost, AI pass-through, and the cost of failures included. It’s the follow-up to our Second Meter investigation, with the meter readings turned into a per-task price list.
The completed task, defined
One task = research one target company and produce one personalized outreach draft, saved and logged. Concretely, the job has these steps:
- Search the web for the company
- Scrape/fetch its site pages
- LLM call: summarize the company into a research brief
- Format/clean the brief
- LLM call: draft the personalized message
- Create/update the CRM record
- Save the draft (doc or email draft)
- Log the run
How each meter reads that identical job:
| Platform | Billing unit | Units consumed by one task |
|---|---|---|
| n8n (Cloud or self-hosted) | 1 execution = the whole workflow run, however many nodes | 1 execution |
| Make | 1 credit per module that executes | ~8 credits (steps 1–3, 5–8 bill; internal formatting may ride along depending on build) |
| Relevance AI | 1 Action per tool run, + Vendor Credits for the AI | ~5 Actions (search, scrape, 2 LLM tool runs, CRM) + vendor credits for both LLM calls |
n8n’s unit is the most forgiving for complex work and the most wasteful for trivial work — a one-node webhook costs the same execution as a 40-node agent. Make’s unit scales with your step count. Relevance’s Action count sits between the two, but it’s the only platform here running a second, separate meter for the AI tokens themselves. Hold that thought; it’s where the 2,000-task table breaks.
The AI pass-through (roughly equal — on purpose)
The two LLM calls in our task — summarize, then draft — use a mini-class model and about 2,000 input / 400 output tokens in total. At published API list prices that’s roughly $0.001–0.003 per task. It is, to a first approximation, the same on all three platforms:
- n8n: you bring your own API key; you pay the provider directly. (n8n Cloud also sells its own AI credits for hosted AI features — Starter includes a monthly AI-credit allowance — but the BYO-key path avoids that meter entirely.)
- Make: same — AI modules can call your own model account, though Make’s built-in AI features draw on your credit bucket, and Make restructured those AI rates in September 2026 (priced in the Make AI credit shake-up).
- Relevance AI: LLM cost is charged as Vendor Credits, passed through at wholesale with no markup per Relevance — 10,000 vendor credits = $20 (so 1 credit = $0.002), included monthly allowance on paid plans, rollover while subscribed, and BYO-key is available on paid plans if you’d rather skip the meter.
At $0.001–0.003 a task, the AI is not the story at this task size. Anyone whose agent bill shocked them wasn’t shocked by a mini-model’s tokens — they were shocked by the platform meter. Which brings us to the platform meter.
Cost per completed task: the tables
Annual-billing plan prices throughout (monthly billing runs ~20–30% higher). Platform cost per task = the monthly bill that volume forces ÷ tasks completed. AI pass-through (~$0.002) excluded from the table and added in the text — it’s a rounding error next to the platform numbers.
At 500 completed tasks/month
| Platform | What you buy | Monthly platform bill | Platform cost / task |
|---|---|---|---|
| n8n, self-hosted | A small VPS (license $0) | ~$6.49 | $0.013 |
| Make Core | $9/mo, 10,000 credits (task uses ~4,000) | $9 | $0.018 |
| Relevance AI Pro | $19/mo, 2,500 Actions (task uses exactly 2,500) | $19 | $0.038 |
| n8n Cloud Starter | $20/mo, 2,500 executions | $20 | $0.040 |
At 2,000 completed tasks/month
| Platform | What you buy | Monthly platform bill | Platform cost / task |
|---|---|---|---|
| n8n, self-hosted | Same VPS | ~$6.49 | $0.003 |
| n8n Cloud Starter | 2,500 executions still covers it | $20 | $0.010 |
| Make Core + extra credits | 16,000 credits needed; Core’s 10,000 + a ~10k credit pack (≈$9–11) | ~$18–20 | ~$0.009–0.010 |
| Relevance AI Pro + Action top-ups | 10,000 Actions needed; 2,500 included + 7,500 top-up at $40–80 per 1,000 | ~$319–619 | ~$0.16–0.31 |
There’s the story. Between 500 and 2,000 tasks a month, n8n’s per-task cost falls 4–13× (the execution unit doesn’t care how big the workflow is, and self-hosting removes the meter altogether), Make’s falls by half (you buy one extra credit pack), and Relevance’s rises 4–8×, because:
- Relevance’s plan ladder doesn’t scale Actions with price. Pro is 2,500 Actions at $19/month billed annually. The next tier, Team, is $234/month billed annually — 12× the price — for 7,000 Actions, not even 3× the allowance, and still short of this workload’s 10,000.
- The overflow path is top-ups, and top-up Actions are priced like a penalty: trackers report $40–80 per 1,000 extra Actions (Relevance’s own pricing has been in flux — its public pricing page showed only an Enterprise “talk to sales” card in September 2026, with self-serve prices living in its docs). At 5 Actions per task, every 1,000 top-up Actions is only 200 tasks.
To be fair to Relevance: its per-Action plan price ($19 ÷ 2,500 = $0.0076) is in Make’s league. The problem is purely the ladder — the jump from “included” to “top-up” pricing, and a Team tier priced for seats and features rather than volume. An agent doing a 2-Action task instead of our 5-Action one fits 1,250 tasks into Pro; task shape decides whether Relevance is reasonably priced or the most expensive option here by an order of magnitude. (n8n’s own agent tooling is evolving fast, too — see our n8n AI Agent Builder preview.)
The failure tax
Completed tasks are what’s left after failures, and each meter treats failures differently:
- Relevance AI: a failed tool run still counts as one Action (per its pricing documentation as reported by trackers). A flaky scraper retrying inside your agent bills you for every attempt.
- Make: credits are consumed as modules execute — a scenario that dies on module 7 of 8 has spent most of its 8 credits, and loops bill per item processed, including the items in a run that ultimately errors.
- n8n: the run is the unit. A 20-node workflow failing at node 19 costs the same 1 execution as a success; the retry costs a second. Failure is cheapest where the unit is coarsest — another structural point for n8n at complexity, and the flip side of its waste on trivial one-step jobs.
Model it honestly: at a 10% failure-and-retry rate, multiply every platform’s per-task figure by roughly 1.1. It doesn’t change the ranking. Nothing at this scale changes the ranking — the meters do.
What this benchmark can’t price
Capability, honestly noted: Relevance’s multi-agent “workforce” model — agents supervising agents, shared knowledge bases, built-in tooling for sales/research workflows — does jobs that would take real assembly in n8n and are awkward in Make. Make’s visual scenario builder remains the fastest way for a non-developer to build a deterministic automation (and its AI agents ride on the same canvas). n8n demands the most technical comfort and returns the most control, including code steps and self-hosting. If your “tasks” are actually multi-agent research projects, Relevance’s per-task premium may be buying something real. If they’re structured jobs like the one defined here, it’s buying meter readings.
FAQ
What is the cheapest way to run 1,000 agent tasks a month?
Self-hosted n8n: the license is $0 for your own business use and a ~$6.49/month VPS covers the volume, putting platform cost under a cent per task plus ~$0.002 of LLM pass-through. Among hosted options, Make Core (~$9/month for the ~8,000 credits our task shape needs) is cheapest, with n8n Cloud Starter ($20/month billed annually) next.
Why is Relevance AI so expensive in the 2,000-task table?
Task shape × plan shape. Our task spends 5 Actions; Pro includes 2,500/month, so 2,000 tasks need 10,000 Actions — and the overflow routes to top-ups at a reported $40–80 per 1,000 Actions, because the Team tier ($234/month billed annually) raises the allowance only to 7,000 while pricing in seats and features. Simpler 2-Action tasks, or volumes under ~500 tasks/month, keep Relevance in its reasonable zone.
Does n8n really charge nothing per step?
Correct — on both Cloud and self-hosted, one workflow execution is one unit regardless of node count. (Cloud plans instead cap executions: Starter’s 2,500/month. And n8n Cloud’s optional built-in AI features have their own AI-credit allowance; bring your own API key and that meter never moves.)
How much do the AI tokens themselves add per task?
About $0.001–0.003 for this task’s two mini-class LLM calls — deliberately modeled small and roughly equal across platforms, since Relevance passes vendor costs through at wholesale and the other two let you bring your own key. Frontier models, long contexts or reasoning-heavy agents change this completely; that’s the second-meter problem from our earlier investigation, and it’s workload-specific, not platform-specific.
Which builder should a solo founder start with?
If your tasks look like the one defined here and volume will grow: n8n (self-hosted if you’re technical, Cloud if you’re not) — its meter is the only one that gets cheaper per task as your workflows get more ambitious. If you want visual building and mostly deterministic flows with some AI steps: Make. If you specifically want managed multi-agent teams and your task volume is modest: Relevance — with your eyes open about the Action ladder.
Can I reduce Relevance costs by simplifying my agents?
Yes — Actions are the lever. Consolidate tool calls, avoid retry loops inside agents (a failed run still bills), and cache research steps outside the agent where you can. Halving Actions per task from 5 to 2–3 roughly halves your effective per-task platform cost and doubles the volume Pro can carry.
Methodology
What we did: Defined one fixed agent task (8 steps, listed in full) and mapped it onto each platform’s billing unit from published pricing documentation: n8n (1 unit per execution), Make (1 credit per executed module), Relevance (1 Action per tool run + Vendor Credits for AI). Priced it at 500 and 2,000 tasks/month using annual-billing plan prices checked in October 2026. Every conversion is shown in the tables.
What we did NOT do: We did not build the agent on any platform, run it, or measure completion quality, speed or reliability. “Cost per completed task” assumes the task completes; platforms differ in how well they do this job, and this article prices only the meter, not the workmanship. LLM pass-through is a directional estimate from published API list prices for a mini-class model at the stated token counts. Relevance’s top-up Action price is reported as a range ($40–80 per 1,000) because current trackers disagree and Relevance’s public pricing page recently showed only its Enterprise tier — we flagged that rather than picking a number.
Conflict disclosure: n8n and Make are affiliate partners of this site; Relevance AI is listed as an affiliate placeholder pending its program terms. None of that changed the tables: n8n wins this benchmark on meter structure, and the article shows the arithmetic so you can check it. Prices checked October 2026.
AWL verdict
Per completed task, the meters decide everything. n8n — self-hosted at ~$0.003–0.013 per task, Cloud at $0.010–0.040 — is the structural winner for agent work at any real volume, because its unit (the whole run) is the only one that doesn’t multiply by steps. Make is the honest middle: ~$0.01–0.02 per task, scaling linearly and predictably with scenario size. Relevance AI is fairly priced inside its Pro allowance (~$0.038/task) and the most expensive option here the moment you outgrow it (~$0.16–0.31/task at 2,000 tasks), because its ladder sells seats, not Actions. Choose the meter whose shape matches your task’s shape — then negotiate price.
Sources
- Relevance AI pricing structure (Free 200 Actions; Pro $29/mo monthly / $19 annual, 2,500 Actions + 10,000 Vendor Credits; Team $349/$234, 7,000 Actions; top-ups $40 per 1,000 Actions, $20 per 10,000 Vendor Credits), via Prospeo pricing analysis, October 2026: https://prospeo.io/s/relevance-ai-pricing-reviews-pros-and-cons
- Relevance AI pricing corroboration — Vendor Credit conversion (10,000 = $20), failed runs count as Actions, pricing-page/Enterprise note (September 2026), via Macha: https://www.getmacha.com/blog/relevance-ai-pricing-explained
- Relevance AI plan table (Pro $19/$29; Team $234/$349; Actions + Vendor Credits model), via ColdIQ, 2026: http://coldiq.com/blog/relevance-ai-pricing
- Make pricing (Free 1,000 credits; Core $9/mo billed annually / $10.59 monthly, 10,000 credits; extra credit packs), via dev.to pricing analysis verified September 2026: https://dev.to/2pizza/make-pricing-explained-what-you-actually-pay-59p3
- Make credit model (renamed operations→credits; per-module billing; September 2026 price cut), via The Plain Manual, October 2026: https://theplainmanual.com/zapier-vs-make/
- n8n Cloud pricing (Starter $24/mo monthly / $20 billed annually, 2,500 executions; Pro $60/$50, 10,000 executions; included AI credits), via ToolQuestor comparison, October 2026: https://toolquestor.com/vs/albato-vs-n8n
- n8n vs Make billing-unit comparison (execution vs operation), via dev.to, October 2026: https://dev.to/stimlau/n8n-vs-make-which-automation-platform-earns-its-keep-3mj8
