Three platforms dominate the automation landscape in 2026, yet their node
catalogs and integration architectures reflect fundamentally different
philosophies. Zapier leads with the largest library —
7,000–8,000+ pre-built app connectors — but charges per task and imposes
a 30‑second timeout on most plans. Make provides 1,500–
2,400 integrations with deeper configuration options per module and a
visual scenario builder, but its polling triggers consume operations even
when no new data exists. n8n ships with 400+ core nodes
plus 600+ verified community nodes and unlimited custom API access via
the HTTP Request node — effectively enabling connections to any service
with a public API [1]
[2].
Critically, n8n counts one execution per entire workflow run regardless
of step count — a structural pricing advantage that compounds with every
node added to a workflow.

How do n8n, Zapier, and Make pricing models differ, and what does this mean at scale?

The pricing model is the single largest cost variable in 2026, far
outweighing the sticker price. Zapier charges per
task — one successful action step. A 10‑step workflow
running 1,000 times a month consumes 10,000 tasks; at that volume the
cost ranges from $99–$200/month. Make charges per
operation (module execution), generally 50–60%
cheaper than Zapier for equivalent volumes, but its polling triggers
burn operations on every interval check even when no new data is found.
[6]
[2]

n8n charges per execution: one complete
workflow run counts as a single unit regardless of whether it contains
5 nodes or 500. This structural difference produces dramatic cost
divergence at scale. A lead enrichment workflow with 5 steps running
5,000 times per month costs approximately $299–$389/month on Zapier
Team, $29–$49/month on Make Pro, and $50/month on n8n Pro; when
self‑hosted on a $5–7/month VPS with the free Community Edition, the
execution cost is zero beyond infrastructure [3].
For enterprises running 25 workflows with 15 steps each at 100
executions per month (37,500 tasks), the three‑year TCO is approximately
$15,600+ for Zapier, $4,200–6,000 for Make, and $4,200–5,000 for n8n
[2]. Polling
triggers compound this further: a Make scenario checking an inbox every
5 minutes burns ~8,640 operations per month on checks alone.
For the complete plan‑by‑plan pricing breakdown, see the
n8n Plans Comparison guide.

Which platforms support custom code execution, and how does n8n’s Code node compare?

n8n is the only platform among the three with a native Code
node — supporting JavaScript (Node.js 18) and
Python for data transformation, JSON parsing, API calls,
and custom logic directly inside the workflow canvas. Zapier provides no
native code execution step — code steps are available only on higher-
tier plans and function as external scripts, not integrated nodes. Make
offers an HTTP module for custom API calls and lets users embed scripts
in its Data Store, but has no native code execution node.
[7]
[3]

This gap is most consequential for AI and data-heavy workflows. When an
LLM returns unstructured or variable JSON, a JavaScript code node can
parse, validate, and reshape the output before it reaches the next node —
something impossible to do natively in Zapier or Make without external
workarounds. n8n also allows self‑hosted users to import external npm
modules (NODE_FUNCTION_ALLOW_EXTERNAL=moment,lodash) and
supports Python’s full standard library when running native Python in
a Task Runner sandbox. For complex data transformation, the Code node
combines with the HTTP Request node to form a universal integration
adapter: any API lacking a native connector becomes accessible without
leaving the workflow canvas. For the complete Code node reference
covering both languages and execution modes, see the
n8n Code Node Transformation guide.

How do AI and LLM node capabilities compare across n8n, Zapier, and Make in 2026?

n8n ships with 70+ native AI nodes built on the
LangChain framework — the most extensive AI integration of any
automation platform. These span Chat Models (OpenAI GPT‑4o, Anthropic
Claude, Google Gemini, Ollama), Embeddings, Vector Stores (Pinecone,
Qdrant, PGVector), Tools, Memory (Window Buffer, Postgres Chat), Chains
(Retrieval QA, Summarization), and the AI Agent node that serves as
the central orchestration layer for tool‑calling agents.
[9]
[10]

In contrast, Zapier launched Zapier Agents in early
2026, enabling autonomous multi‑step task execution with an AI Copilot
that builds Zaps from natural language descriptions. However, a March
2026 Stack AI analysis found Zapier’s AI capabilities remain “basic
capabilities based on triggers and actions” compared to purpose‑built
AI platforms. Make introduced its Maia AI assistant
and Make AI Agents for autonomous task execution, offering native
integrations for OpenAI, Anthropic Claude, and Google AI with
templated prompt inputs. The core architectural distinction: n8n’s
LangChain foundation supports persistent agent memory, vector database
RAG, local LLM execution (Ollama) for regulated industries, and no
execution time limits when self‑hosted — LLM chains running for minutes
are fully supported. Zapier’s AI operates within a 30‑second timeout
on most plans; Make’s timeout is longer but still capped. For the
complete AI agent orchestration guide, see the
n8n AI Nodes reference.

How does self‑hosting and open‑source access differ between these platforms?

Self‑hosting is the most binary differentiator among the three platforms.
n8n is the only platform that supports self‑hosting:
you can deploy it on your own infrastructure — a $5–7/month VPS, a
Docker‑managed server, or on‑premises hardware — using the free
Community Edition under the Sustainable Use License. This grants
unlimited workflows, zero per‑execution charges, full data sovereignty,
and the ability to connect to services running inside your private
network. Zapier and Make are both
cloud‑only SaaS products; you cannot host either on your own servers.
[7]
[8]

The Sustainable Use License is “fair‑code” — not OSI‑approved open
source — meaning the source code is always visible and self‑hosting
is always permitted, but commercial redistribution or embedding the
platform itself requires a paid license. For regulated industries —
healthcare, finance, government — n8n’s self‑hosted option is the
only path to meet data‑residency requirements without a custom
enterprise agreement from a SaaS vendor. N8n also incurs no
per‑software cost when self‑hosted, but DevOps time (roughly 1–3 hours/
month for updates and maintenance) must be factored into the total
cost calculation. For the complete self‑hosted hardening blueprint
covering Nginx reverse proxy, TLS termination, firewall rules, and
Basic Auth, see the
n8n Node Security Hardening guide.

What does total cost of ownership look like for a mid‑sized automation workload?

For a team running 10 active workflows at 50,000 total workflow runs
per month, the numbers diverge sharply. An n8n self‑hosted instance on
a $5–20/month VPS handles this volume with zero software licensing
fees; n8n Cloud Pro covers it for approximately $50–65/month because
a 10‑step workflow running 10,000 times counts as just 10,000
executions. Zapier Team for the equivalent would reach $400–600/month
(the same 10‑step workflow billed as 500,000 tasks for 50,000 runs).
[11]
[5]

Make’s Pro plan hovers around $150–250/month for this workload. For
enterprises running 150 processes with 25 workflows, three‑year TCO
is approximately $1,200 for a self‑hosted n8n scenario (server costs
and light maintenance) versus $10,000–15,000 for Zapier or Make when
factoring in license, usage, and team training [2].
The fundamental driver is pricing architecture: Zapier’s per‑task and
Make’s per‑operation models scale linearly with complexity and volume;
n8n’s per‑execution model is flat regardless of complexity. For the
complete TCO calculation framework covering all plan tiers,
infrastructure costs, and team sizing, see the
n8n Plans Comparison guide.

Which platform has the best node catalog when extensibility and AI are your top priorities?

The answer depends entirely on your team profile. For large
integration breadth with zero technical effort, Zapier’s
7,000–8,000+ pre‑built connectors are unmatched — if the app exists in
the SaaS mainstream, Zapier likely has it. For visual builders
who need moderate depth with predictable pricing, Make’s
1,500–2,400 integrations with deep configuration options per module
offer the best middle ground. Its visual scenario builder is superior
to Zapier for multi‑branch logic and data transformation.
[5]
[7]

For developers, agencies, and AI‑heavy teams, n8n’s
catalog is effectively the largest because the 400+ core nodes and
600+ community nodes represent the built‑in surface area, while the
HTTP Request node and Code node connect to anything with an API —
making the effective integration count unlimited for technical teams.
The 70+ LangChain AI nodes, persistent agent memory, vector store RAG,
local LLM support (Ollama), and unlimited execution time when self‑
hosted make n8n the clear choice for AI‑centric automation in 2026.
The decision rule: choose Zapier for non‑technical
teams needing simple linear automations under 2,000 tasks/month with
guaranteed uptime; choose Make for ops teams who
want powerful visual branching at moderate volume; choose
n8n when cost at scale, code‑level extensibility, or data
residency through self‑hosting are non‑negotiable. For the complete
platform selection framework covering data sovereignty and compliance
requirements, see the
n8n Architecture & Scaling Hub.