
For agents that call several apps and need guardrails, Zapier’s AI step or Make is the lowest-friction start. n8n is the better pick once runs get long or volume gets high, because it bills one flat unit per run. Lindy suits personal-assistant work, and Relevance AI is no longer a self-serve option for small teams.
A workflow is a straight line. An agent bills a loop. The model reads the goal, picks a tool, reads the result, and decides again, and every pass through that loop can be metered differently. Most comparison pages rank these platforms by integration count and template libraries. The choice that decides your bill is the billing unit, and that is what this article ranks them by.
One disclosure first. This comparison rests on vendor pricing pages, vendor documentation, and third-party checks, all read in November 2026. Nothing here comes from hands-on testing, and every cost figure marked “calculation” is editorial arithmetic on published rates. Sources disagree more than you would expect, so the last section ranks them by reliability.
One loop, five different meters
The diagram summarizes each vendor’s published billing rule. n8n states that pricing is based on monthly workflow executions, regardless of complexity, so a run with two steps and a run with forty cost the same. Make charges 1 credit per operation, plus token credits when you use Make’s own AI provider, and its agents are still in open beta. Zapier says AI steps bill as tasks, and the model tier sets a multiplier of 1, 3, or 5. Lindy’s rule is that credits are consumed by everything the agent does after being triggered. Relevance AI charges Actions for each tool run and Vendor Credits for model costs passed through at wholesale. www.withorb.com +4
The consequence is simple. On a flat-per-run meter (n8n), a chatty agent costs the same as a quiet one, and you pay your model provider separately for tokens. On a per-call meter (Zapier, Relevance AI), every extra tool call raises the bill. Make sits in between, and Lindy bills per step without publishing a per-run figure I could verify.
The five platforms, with a verdict each
: n8n: best for control and predictable cost
n8n’s Starter plan is 20 € a month billed annually, with 2.5K executions. Self-hosting is possible too: the free Community Edition runs under the Sustainable Use License v1.0, so read the terms before you build a product on it. n8nanavem
Its strongest feature for agents is approval control. n8n’s docs let you require human approval before an AI Agent executes a specific tool. The workflow pauses, a reviewer sees which tool the agent wants and with what parameters, and a denied call is canceled. That is a per-tool brake, which is the right place for one. n8n
The catch is the “no code” label. n8n is a visual canvas, but you will meet expressions, credentials, and prompts that must describe your approval setup. The docs warn that the system prompt needs the tool setup and review steps spelled out. A non-technical operator can get there, but not in an afternoon.
Zapier: easiest start, hardest bill at volume
The most important recent fact is that standalone Zapier Agents is gone as a separate product. Zapier’s October 6, 2026 announcement says agent capabilities now live inside a single AI by Zapier step in the Zap editor. Its migration guide adds that you no longer manage a separate Agents subscription or add-on fees. zapierzapier
Many pricing pages still quote the old “activities” model. Treat those as stale for planning.
The upside is real: your agent now sits inside a Zap, so it can use filters, paths, and run history, and Zapier says it can start from any of 9,000+ app integrations. Zapier also built in a brake: a run that crosses 75 tasks pauses for a human instead of looping quietly. createwithzapier
The downside is the meter. A third-party reading of Zapier’s documentation gives the formula tasks per run = (1 × model rate) + (tool calls × model rate). A Premium-tier step with two tool calls therefore costs 15 tasks. Also, full-Zap tests consume tasks the same way production runs do, so testing is not free. If you migrate an existing agent and forget to switch the old one off, both the Zap and the agent can run at the same time and cause duplicate actions. zapier agents to ai by zapier +2
Make: best price for visual builders, with a beta warning
Make agents are built inside the scenario builder, the same canvas you use for ordinary automations. The agent app has been in open beta since February 2, 2026, and Make says functionality and pricing may change. Paid plans can bring their own OpenAI or Anthropic key, in which case Make charges the operation credit and your provider bills the tokens. latenode
Make’s own advice is to give agents only tasks you would trust an intern with, and to avoid sensitive data and high-stakes financial decisions. I would add one operational point from the same source. Paid plans can auto-purchase credits at a 25% premium, so an agent stuck in a retry loop can keep spending instead of stopping. For unattended agents, turn auto-purchase off and set the steps-per-call ceiling before the first live run. latenode
Lindy: right for a personal assistant, wrong for pipelines
Lindy has moved toward being an assistant that lives in your messages. As of late October 2026 its pricing page lists $29.99 a month per seat for 3,000 pooled credits, with top-ups at $10 per 1,000. There is no permanent free tier: free credits last 7 days. duppleanavem
Its own pages don’t agree with each other. docs.lindy.ai shows Pro at $100 a month while the docs billing page says $99.99, and the plan names differ from lindy.ai/pricing. Older reviews still quote a $49.99 entry plan. Use it for inbox triage, scheduling, and follow-ups. Don’t pick it to run a high-volume backend flow, because a pooled-credit seat model rewards low volume per person. anavem
Relevance AI: not a self-serve choice for small teams anymore
Relevance AI split pricing into Actions and Vendor Credits in September 2025, but the public page is now the problem. Several independent checks in September and October 2026 found that its pricing page has one plan, Enterprise, priced by sales. The documentation still lists Pro and Team tiers, and one check read them as 2,500 Actions and 10,000 Vendor Credits for Pro, and 7,000 and 35,000 for Team. Top-ups run $80 per 1,000 Actions and $20 per 10,000 Vendor Credits. Relevance AI: What It Is, Pricing, and Alternatives (2026) +2
When a vendor’s public page and its own docs describe different buying paths, plan on a sales conversation. For a solo builder or a five-person team, that disqualifies it from a quick comparison.
The same job on each meter (calculation)
Take one agent run: a trigger, two tool calls, and roughly 6,000 input and 800 output tokens across the model calls. The token figures are an assumption from a worked example in a third-party Make guide. Substitute your own measured numbers.
| Platform | Billable units for this run | Approx. cost at entry tier |
|---|---|---|
| Make (Large tier, Make’s AI provider) | ~7.4 credits | ~$0.007 |
| Zapier (Standard, 1×) | 3 tasks | ~$0.08 |
| Zapier (Premium, 5×) | 15 tasks | ~$0.40 |
| n8n (Starter) | 1 execution | ~€0.008, plus your own token bill |
| Relevance AI (top-up rate) | 2 Actions | ~$0.16, plus Vendor Credits |
The inputs are these. Make Core is $9 a month for a 10,000-credit minimum, which is $0.0009 per credit, and Make’s published rate table gives 3,616 input and 452 output tokens per credit on the Large tier. That makes 4 operation credits, plus 1.66 for input and 1.77 for output tokens. Zapier Professional’s entry tier is $19.99 a month billed annually for 750 tasks, about $0.027 a task. n8n is €20 divided by 2,500 executions. make ai agents +2
Three cautions apply. These are entry-tier prices, and per-unit cost falls on higher volume tiers. Zapier’s “Standard” and Make’s “Large” are different models, so the gap reflects billing design as well as model choice. And n8n’s figure excludes tokens entirely, so it is not a like-for-like total. What the table does show is direction: for a tool-heavy agent, the per-call meters grow with every tool call, and the per-run meter does not.
What to do differently after reading this
- Count tool calls per run before you pick a platform. At two calls, the gap is a few cents. At ten calls per run on Zapier Premium, one run is 55 tasks, close to the 75-task pause. If your design needs that many calls, price it on n8n or Make.
- Let agents read and draft on their own; require approval for sends, payments, and deletions. n8n gives you tool-level approval in the docs, and Make’s guidance points the same way. Put the hard rule in the tool, such as a filter that only lets a refund through below your limit, because instructions in a prompt can be ignored.
- Budget for tests. Zapier counts full-Zap tests as tasks, and Make charges credits for chatting with an agent. Run your 20-run test on a free or entry plan with a spend alert.
- Make writes idempotently. Make’s MCP documentation says a call that times out can keep running, so a retrying client can run it twice. Search for an existing record before creating one.
- Skip Relevance AI unless a sales process suits you, and skip Lindy for pipelines.
The question this comparison cannot answer yet
Which platform has the lowest real cost for your agent depends on measured tokens per run, and nobody publishes that. Make’s own worked example prices a single AI step. It does not price a full run. The data that would settle it is 50 real runs on your workload, logged with credits or tasks consumed. Until then, treat any “cheapest platform” ranking as a hypothesis, including this one.
A second open question is whether Make’s beta pricing holds. Make says it may change, so recheck its credit rates when the agent app leaves beta.
Which sources to trust, and where they conflict
Strongest: vendor primary material, namely n8n’s pricing page and docs, Zapier’s October 6, 2026 announcement and help guide, and Relevance AI’s docs.
Medium: third-party summaries of vendor documentation (Latenode, Anavem, Dupple, Toolradar). Latenode publishes a competing platform, though it cites Make’s help pages throughout.
Weakest: pricing aggregators, which contradict each other:
- n8n’s Business tier appears as 667 € a month on one site, $800 on another, and $960 after a May 2026 increase on a third. I left it out for that reason. www.anavem.com +2
- Zapier Agents Pro on monthly billing is listed at about $40 on one page and $50 on another. pageszapier
- Lindy’s entry price appears as $49.99 in older pages and $29.99 in current ones.
Next step: pick one real workflow, build it on your two shortlisted platforms with a spend alert, and log credits or tasks per run across 50 runs. That number beats any table in this article.
