Zapier vs Make vs n8n: Real AI Automation Cost per 10,000 Workflow Runs in 2026
The same AI workflow can consume radically different billing units on these three platforms, and not just because of how many steps it has. Since June 15, 2026, Zapier's AI steps carry a task multiplier that depends on which model tier is selected and how many tools that step calls — a Premium-tier AI step with two tool calls costs 15 tasks by itself, while the identical logic run through a bring-your-own API key costs 5. Make's credit system has a parallel split: an AI module using your own connected OpenAI or Anthropic account bills Make at a flat 1 credit, with tokens paid directly to the provider, while the same call routed through Make's own AI Provider bills a dynamic, token-based credit amount that varies by model and prompt length. n8n remains the simplest of the three: one execution per run, regardless of steps, tools, or which AI provider is used.
The short answer
For a simple, tool-free AI step, all three platforms are inexpensive at any volume modeled. For a tool-using workflow, the model tier chosen on Zapier is now as important as the run volume itself: the same logic costs roughly 6 tasks per run on Zapier's bring-your-own-key option, 16 on its mid-tier Advanced model, and 26 on the Premium tier that Zapier sets as the default for new AI steps — a more than fourfold spread from model choice alone, before volume is even considered. Make's cost depends similarly on connection type: routing AI calls through your own provider connection keeps Make's own credit consumption flat and predictable, while using Make's built-in AI Provider adds a token-based, prompt-length-sensitive credit cost on top. n8n's per-run billing is unaffected by any of this, which is precisely why the gap between n8n and the other two widens specifically as workflows get more tool-heavy and providers get pricier, not just as run volume grows.
How each platform actually bills, as of the current model
| Zapier | Make | n8n | |
|---|---|---|---|
| Billing unit | Task — for AI steps specifically, governed by a model-tier multiplier since June 15, 2026 | Credit — replaced "operations" as Make's billing currency; most non-AI modules still convert 1:1 | Execution — one entire workflow run, regardless of node count or which AI provider is used |
| AI billing formula (Zapier) | Tasks per AI step run = (1 × model rate) + (tool calls × model rate), where model rate is 1x (Standard, no tools), 3x (Advanced, tools supported), 5x (Premium, tools supported, the default for new steps), or 1x (bring your own API key, tools supported with Advanced/Premium models) | Not applicable in the same form | Not applicable — flat per-run regardless of model or tool use |
| AI billing formula (Make) | Not applicable | Third-party AI app with your own connection (OpenAI, Anthropic, Gemini): 1 operation = 1 credit on Make's side; you pay the provider directly for tokens. Make's own AI Provider (used in Make AI Agents and Make AI Toolkit): dynamic credits based on both tokens and operations, varying by the specific model selected, with input and output tokens converted to credits at different rates | Not applicable |
| Free tier | 100 tasks/month, 2-step Zaps only | 1,000 credits/month, 2 active scenarios | No perpetual free cloud tier; Community Edition (self-hosted) is free and unlimited |
| Entry paid tier | Professional: $19.99/month (annual), 750 tasks | Core: $9–12/month (annual), 10,000 credits | Cloud Starter: $20–24/month, 2,500 executions |
| Mid tier | Team: $49–69/month, 2,000 tasks | Teams/Pro: $16–29/month, 10,000–40,000 credits depending on tier | Cloud Pro: $50/month (about €50), 10,000 executions |
| High-volume published tier | Reported around $733.50/month (annual) for 100,000 tasks | Higher tiers scale to roughly $99–175/month for 80,000–150,000 credits | Beyond Pro, pricing is not published; self-hosted removes the execution cap entirely |
| Self-hosting | Not available | Not available | Available; Community Edition unlimited, cost is your own server (reported $5–15/month for a small VPS) |
| Non-AI action cost | 1 task per app action, unaffected by the AI multiplier | 1 credit per module run for the large majority of non-AI apps | Included in the flat per-run execution count |
The Zapier task formula, with confirmed examples
Zapier's own documentation states the formula plainly: tasks used per AI step = (1 × model rate) + (tool calls × model rate). Zapier's own worked examples: a Premium-tier step with 2 tool calls costs 15 tasks; an Advanced-tier step with 1 tool call costs 6 tasks; a bring-your-own-key step with 4 tool calls costs 5 tasks. New AI steps default to Premium (5x) unless a lower tier is explicitly selected, which means a team that doesn't actively manage model tier is very likely paying the most expensive multiplier by default. Legacy AI steps created before June 15, 2026 are frozen at their old behavior and cannot use tools at all — as of June 30, 2026, legacy steps using knowledge sources on a Standard model were paused entirely, and creating tool-using workflows now requires building a new-style AI step with an explicit tier selection.
The Make credit split, with confirmed rates
Make's help documentation is explicit that credit usage depends on connection type, not just which app is used. A third-party AI app module using your own account (OpenAI, Anthropic Claude, Gemini) bills Make at a fixed 1 credit per operation, with token cost paid separately, directly to that provider — Make's own bill stays predictable regardless of prompt length. Make's own AI Provider, by contrast, uses a published, model-specific token-to-credit conversion table with separate rates for input and output tokens — as one confirmed example, Anthropic Claude Sonnet 4.5 through Make's AI Provider converts at 301 input tokens per credit and 60 output tokens per credit, meaning longer prompts and longer generated responses cost more credits in a way that a fixed 1-credit third-party-connection module never does.
Three representative AI workflows, rebuilt on current billing rules
Illustrative step counts and token estimates, not vendor-defined benchmarks:
- A. Simple: trigger → one AI call (no tools) → one app action.
- B. Tool-using: trigger → AI reasoning orchestrating 4 tool calls → database write (5 total steps beyond the trigger).
- C. Larger agentic: trigger → 3 AI reasoning calls, each with 2 tool calls → 10 further downstream operations.
Zapier task counts, by model tier
| Workflow | Standard (1x) | Advanced (3x) | Premium (5x, default) | Bring your own key (1x) |
|---|---|---|---|---|
| A (1 AI call, 0 tools, + 1 action) | (1×1)+0 tasks for the AI step + 1 for the action = 2 tasks | Not typically needed (no tools) | (1×5)+0 + 1 action = 6 tasks | (1×1)+0 + 1 action = 2 tasks |
| B (AI orchestrating 4 tools + 1 db write) | Not usable — Standard doesn't support tools | (1×3)+(4×3) + 1 = 3+12+1 = 16 tasks | (1×5)+(4×5) + 1 = 5+20+1 = 26 tasks | (1×1)+(4×1) + 1 = 1+4+1 = 6 tasks |
| C (3 AI calls × 2 tools each + 10 downstream ops) | Not usable — Standard doesn't support tools | 3 × [(1×3)+(2×3)] + 10 = 3×9+10 = 37 tasks | 3 × [(1×5)+(2×5)] + 10 = 3×15+10 = 55 tasks | 3 × [(1×1)+(2×1)] + 10 = 3×3+10 = 19 tasks |
Cost at 1,000, 10,000, and 100,000 runs
| Runs/month | Workflow, tier | Total tasks | Zapier cost (nearest published tier) |
|---|---|---|---|
| 1,000 | A, Standard/BYO (2 tasks/run) | 2,000 | Team tier, $69/month |
| 1,000 | B, bring-your-own-key (6 tasks/run) | 6,000 | Team tier's 10,000-task band, $129/month |
| 1,000 | B, Premium default (26 tasks/run) | 26,000 | Reported ~100,000-task tier, ~$733.50/month |
| 1,000 | C, bring-your-own-key (19 tasks/run) | 19,000 | Reported ~100,000-task tier, ~$733.50/month |
| 1,000 | C, Premium default (55 tasks/run) | 55,000 | Reported ~100,000-task tier, ~$733.50/month |
| 10,000 | A, Standard/BYO (2 tasks/run) | 20,000 | Reported ~100,000-task tier, ~$733.50/month |
| 10,000 | B, bring-your-own-key (6 tasks/run) | 60,000 | Reported ~100,000-task tier, ~$733.50/month |
| 10,000 | B, Premium default (26 tasks/run) | 260,000 | Above every published tier — custom/Enterprise |
| 10,000 | C, bring-your-own-key (19 tasks/run) | 190,000 | Above every published tier — custom/Enterprise |
| 10,000 | C, Premium default (55 tasks/run) | 550,000 | Above every published tier — custom/Enterprise |
| 100,000 | Any workflow above, any tier | 200,000–5,500,000 | Every combination exceeds published tiers — custom/Enterprise |
The single most consequential number in this table isn't a volume tier — it's the ratio between Workflow B's bring-your-own-key task count (6) and its Premium-default task count (26): more than 4x, entirely from a model-tier setting most teams never touch, since Premium is Zapier's own default for every newly created AI step.
Make credit counts, by connection type
Using the published Claude Sonnet 4.5 conversion rate (301 input tokens/credit, 60 output tokens/credit) as a representative Make AI Provider example, and illustrative token counts per call (workflow A: 500 input/100 output tokens; workflow B and each call in C: 1,500 input/300 output tokens):
| Workflow | Own-connection credits (1 op = 1 credit; tokens billed by provider directly) | Make AI Provider credits (dynamic, tokens + operations) |
|---|---|---|
| A (1 AI call + 1 action) | 2 credits | (500÷301) + (100÷60) + 1 action ≈ 4.33 credits |
| B (AI + 4 tool ops + 1 db write) | 6 credits | (1,500÷301) + (300÷60) + 5 other ops ≈ 14.98 credits |
| C (3 AI calls + 6 tool ops + 10 downstream ops) | 19 credits | 3 × [(1,500÷301)+(300÷60)] + 16 other ops ≈ 45.95 credits |
Cost at 1,000, 10,000, and 100,000 runs, at an illustrative $0.001/credit blended rate
| Runs/month | Workflow, connection type | Total credits | Illustrative cost |
|---|---|---|---|
| 1,000 | A, own connection | 2,000 | $2.00 |
| 1,000 | A, Make AI Provider | 4,330 | $4.33 |
| 1,000 | C, own connection | 19,000 | $19.00 |
| 1,000 | C, Make AI Provider | 45,950 | $45.95 |
| 10,000 | A, own connection | 20,000 | $20.00 |
| 10,000 | C, own connection | 190,000 | $190.00 |
| 10,000 | C, Make AI Provider | 459,500 | $459.50 |
| 100,000 | A, own connection | 200,000 | $200.00 |
| 100,000 | C, own connection | 1,900,000 | $1,900.00 |
| 100,000 | C, Make AI Provider | 4,595,020 | $4,595.02 |
The $0.001/credit rate is a blended illustration based on Make's published Core-tier price range; actual per-credit cost varies by plan tier and volume commitment, and Make's own AI Provider consumption depends heavily on prompt and response length, which will differ from the illustrative token counts used here for any real workflow.
n8n stays flat regardless of any of this
At every combination of workflow, model tier, and connection type shown above, n8n's cost is simply the run count against its execution-based tiers: 1,000 runs fits Cloud Starter ($20–24/month) at any of the three workflow shapes; 10,000 runs fits Cloud Pro ($50/month); 100,000 runs exceeds n8n Cloud's published tiers and requires Enterprise, the same as the other two platforms at that volume. Self-hosted n8n's cost (reported $5–15/month for a small VPS) doesn't move at all across this entire table, because it was never counting tasks, credits, tokens, or tool calls in the first place.
What the advertised price misses
- Zapier's Premium default is easy to miss and expensive when missed. A team that never explicitly sets the model tier on a new AI step is billed at 5x, the highest multiplier available — deliberately checking and, where output quality allows, downgrading to Advanced (3x) or a bring-your-own-key connection (1x) is a real, immediate lever with no code change required.
- Zapier's 75-task-per-run pause limit is a real operational constraint, not just a cost one — if a single AI step's run reaches 75 tasks (a Premium-tier step with roughly 14 tool calls, by the formula), Zapier pauses it and requires manual approval before continuing, which can silently stall a complex agentic workflow rather than simply billing more.
- Make's own AI Provider is more expensive to reason about than a fixed per-operation rate, precisely because its cost depends on prompt and response length in a way that a company's own OpenAI or Anthropic connection doesn't add to Make's bill at all — a team optimizing Make cost should generally prefer bringing its own AI provider connection over Make's built-in one, accepting the provider's own token billing as a separate, more transparent cost.
- Legacy Zapier AI steps (created before June 15, 2026) cannot use tools and had some configurations paused entirely on June 30, 2026 — any workflow relying on an older AI step needs to be rebuilt as a new-style step with an explicit model tier before it can use tools, an unavoidable one-time migration cost the per-task pricing itself doesn't reflect.
Break-even: when does self-hosting n8n win?
Self-hosted n8n's cost is essentially flat (server cost only) regardless of AI model tier, tool count, or connection type, while n8n Cloud, Zapier, and Make all scale with those factors. The crossover depends on how much a team values not managing infrastructure: at 10,000 runs of a Premium-tier, tool-heavy Zapier workflow (custom/Enterprise pricing) or a Make-AI-Provider-heavy workflow ($459.50/month, illustrative), self-hosted n8n's flat $5–15/month is decisively cheaper — the crossover in n8n's favor happens earlier and more sharply for AI-heavy, tool-using workflows than it does for simple ones, precisely because Zapier's and Make's AI-specific billing (unlike their base per-step or per-operation rates) has no equivalent on a platform that doesn't meter tokens or tool calls at all.
Which platform fits which workflow
- Simple, tool-free AI steps at any volume: any of the three; Zapier's larger app catalog (9,000+ integrations) may matter more than the small cost difference at this workflow shape.
- Tool-using AI workflows on Zapier specifically: actively choose the model tier rather than accepting the Premium default — the difference between bring-your-own-key and Premium is worth checking on every AI step before it goes to production.
- AI-heavy workflows on Make: prefer your own AI provider connection over Make's built-in AI Provider wherever output quality allows, to keep Make's own credit consumption flat and predictable and let the provider's own transparent token pricing carry the AI-specific cost instead.
- High-volume, tool-heavy, or multi-model-call agentic workflows: n8n, whose flat execution-based billing is unaffected by every one of the multipliers that make Zapier's and Make's AI-specific costs hard to predict in advance.
Limitations and uncertainty
Zapier's AI model-tier task formula and its worked examples are confirmed directly from Zapier's own help documentation, current as of an August 2026 update. Make's credit-versus-connection-type billing logic and its published Claude Sonnet 4.5 token-to-credit conversion rate are confirmed directly from Make's own help documentation, current as of a September 25, 2026 update. Zapier's and Make's published tier prices beyond their entry plans (the ~$733.50/month Zapier figure and Make's higher credit tiers) are reported by third-party trackers rather than confirmed directly against a live vendor page. The specific token counts assumed for each illustrative workflow, and the $0.001/credit blended rate used for Make's dollar figures, are this article's own assumptions, not vendor-published benchmarks — a real workflow's actual token consumption will vary by prompt design and model response length.
Sources
Checked late September 2026. Zapier: zapier.com/pricing, help.zapier.com's "AI by Zapier model tier pricing" article (updated August 20, 2026) and its "AI by Zapier new model-based pricing starting June 15, 2026" announcement. Make: make.com/pricing, help.make.com/credits (updated September 25, 2026) and Make's AI Provider token-pricing update documentation. n8n: n8n.io/pricing. Tier prices beyond each vendor's entry plan are reported by multiple independent third-party pricing trackers.