Make, n8n or a custom API: choosing an AI workflow platform

· Make (cloud iPaaS): a fully managed visual platform where a Scenario is a trigger plus a sequence of modules; it offers a REST-style open API and a custom-app mechanism, and most administrative endpoints require a paid subscription (see the Make developer documentation, verified 2026-10-28);
· n8n (fair-code automation tool): defined in its official documentation as a "fair-code licensed workflow automation tool"; you can install it yourself (the documentation lists Docker Compose and other deployment options) or use its cloud service (see the n8n documentation, verified 2026-10-28);
· Custom API: orchestrate model calls directly in code (queue + functions + scheduled jobs), with no platform layer — all the control and all the responsibility.

Make (USD, page list prices): Free $0/month (explicitly including 1,000 credits/month); Core $9 / Pro $16 / Teams $29 per month (these numbers are shown in a "Save 15%" context, and the page does not state whether they are monthly or annual billing; side-by-side monthly/annual pricing is not given); Enterprise is contact-sales. When credits run out: scenarios stop running; you can upgrade, buy add-ons (only 1,000/10,000 bundles at "a fixed price defined in the subscription" — the number is not public) or enable auto-purchase of 10,000 credits on Core/Pro/Teams; credits expire each cycle and do not roll over. What a credit counts: most module actions in a scenario count as 1 credit each (read/search/create/update/transform/aggregate/iterate all count); error handlers and Router modules do not; Make Code is billed separately at 2 credits/second; the vendor says a single scenario run consumes "from 2 credits to thousands". Source: Make official pricing page (second verification 2026-10-28).

n8n (EUR, page list prices): Starter €20/month (annual, includes 2,500 workflow executions/month); Pro €50/month (annual, 10,000 executions/month); Business €667/month (annual, 40,000 executions/month, self-hosted only); Community edition is free and self-hosted (via GitHub). What an execution counts: one execution = the whole workflow completing once, regardless of how many steps or how much data it processes. Source: n8n official pricing page.

The two units are not interchangeable: one Make credit ≈ one module action, while one n8n execution = one full run (which may contain dozens of modules). Comparing "10,000 credits" with "10,000 executions" directly is wrong — the same flow can consume an order of magnitude differently on each platform. Items not published (not guessed): how many credits each Make paid tier includes, monthly billing prices and add-on bundle prices; n8n monthly billing prices, overage rates and the cloud Business launch date.

Three layers: ① platform subscription/usage (Make credits or n8n cloud executions; with self-hosting, ② infrastructure — servers and ops labor); ③ LLM API fees (identical across all three — see list prices in the model library). Compute the layers separately instead of asking "which total is lowest".

Example workflow (illustrative assumptions, not measured statistics): 500 runs/day of "form → AI tagging → write to sheet", each with one model call (2,000 input / 800 output tokens, 30% cache hit rate). 15,000 runs a month:

· Make side (assuming 5 credits per run: trigger + fetch + trim + HTTP + write): 15,000 × 5 = 75,000 credits/month. Included quotas per tier are not public, so no official plan combination can be priced — the following is a model estimate (not an official quote): linear extrapolation at the publicly shown Pro unit price of $1.60 per 1,000 credits gives 75,000 × $1.60 ÷ 1,000 ≈ $120/month. Basis and limits: ① subscription-included quotas not counted (not public); ② the official path is "subscription + 1,000/10,000 add-on bundles" at non-public fixed prices, so linear extrapolation exceeds the official bundling rules; ③ the actual monthly fee must follow Make's official quote/console.
· n8n cloud route: 1 run = 1 execution → 15,000 executions/month. Pro at €50/month (annual) includes only 10K, so it is exceeded; the vendor says over-quota usage does not stop service, but Starter/Pro overage rates are listed only as "Capped" with numbers not public — the exact cloud monthly fee cannot be determined from public pages; cloud Business is not yet available (waitlist only).
· n8n self-hosting route: Business at €667/month annual (40K executions/month; the license key must connect to n8n's license server daily) + your own server and operations cost (sizing undecided, not estimated). Note €667 is a self-hosted plan price, not a cloud price — the two are not interchangeable.
· LLM API side (identical for all three): at DeepSeek V4.1 Flash off-peak rates ≈ ¥7/month (15,000 runs × ≈¥0.00046 per run; list-price estimate from models.json, verified 2026-10-28);
· How to read it: at this volume, platform/infrastructure cost (tens to hundreds of USD, with the Make side estimable only) dwarfs the API fee (single-digit CNY) — optimize "flow complexity and run count" before obsessing over model unit prices. One order of magnitude smaller, Make's free tier (1,000 credits/month) or n8n's trial may be enough outright.

Limitations (labeled item by item): 5 credits/run is an illustrative assumption (official range: 2 to thousands per run); Make's per-tier included quotas, add-on fixed prices and monthly billing prices are not public; n8n monthly billing prices and overage numbers are not public; self-hosting infrastructure is not estimated. These are worked examples from documentation research — not industry averages, not measured bills; pricing follows official sites, verified 2026-10-28.

· Small team validating fast: a visual platform is quicker; prefer fully managed when business data is not sensitive (one less ops chore);
· Data must not leave the intranet: self-hosted n8n or a custom API; neither hands business data to a third-party platform;
· High volume with ops capability: self-hosting is usually cheapest overall (no per-operation fees), but rate limiting, retries, monitoring and alerting become your own job — they come built-in on platforms;
· Flows change frequently: visual tools rewire links; custom APIs need code changes plus redeployment;
· Not sure yet: trial at low volume on a platform and measure real token consumption (the architecture section of the Make workflow cost guide generalizes to all three options), then decide whether to migrate.

Model APIs are the shared raw material of all three options, priced by the vendors (the list prices in this site's models.json, each individually verifiable). What a platform decision should really compare: platform fees, operations labor, data boundary and lock-in cost — not "which option gets cheaper model calls": none does.

· Commercial relationship: Make's official affiliate program is in candidate evaluation on this site — not applied for, not approved — and this article contains no promotional links; n8n and the custom-API option have no commercial relationship with this site. If Make is ever approved, the CTA will appear only near this section with the commission relationship disclosed, and conclusions will not change;
· Sources: conceptual descriptions of n8n and Make quote their official documentation (verified 2026-10-28, links above); both vendors' pricing was verified and unverified items are labeled;
· This article is documentation research with no hands-on conclusions; no click or conversion data exists (this site configures no such analytics).

· Estimate the shared LLM cost for all three options: cost calculator;
· Platform workflow cost structure: Make workflow cost guide;
· Model prices and comparisons: model library / model comparison;
· Back to the guide directory.