How much does an AI step cost in Make credits?
It depends on the model more than anything else. A step that reads 2,000 tokens and writes 300 costs about 1.2 credits on GPT-5 nano, 8.8 on Claude Sonnet 5 and 39.9 on Claude Fable 5, including the one credit for the operation.
We built this dataset from Make's own token-to-credit table, read October 30, 2026, and converted every row into credits and dollars per million tokens. Then we checked the ten Claude rows against Anthropic's published API prices. The result: at $0.0009 a credit, Make's AI Provider charges exactly Anthropic's list price for Claude tokens, so your plan's per-credit price decides whether Make's built-in AI costs more or less than your own API key. For how we build Make scenarios with AI steps, see our Make.com automation service.
How did we build this dataset?
We copied all 30 rows of Make's published tokens-per-credit table, divided one million by each rate, and priced the result at the four per-credit prices Make quotes in its own Help Center. Every input is a primary page read October 30, 2026.
- Token rates. Make's credits guide lists, for Make's AI Provider, how many input tokens and how many output tokens one credit buys: 3 tier rows (Small, Medium, Large) and 27 model rows from OpenAI and Anthropic. The page was last updated October 29, 2026.
- Credit prices. Make's pricing page does not print plan prices as plain text, so we used the prices Make states in its Help Center: a $9 plan with 10,000 credits ($0.0009 a credit) in the extra credits guide, Core at $113.85 for 150,000 credits ($0.000759) in the change your subscription guide, $338.13 for 500,000 credits ($0.00067626) in the plans and pricing note, and extra credits at 25% more ($0.001125 on the $9 plan).
- Direct prices. Anthropic's API pricing page for base input and output tokens, read the same day.
- Formula. Credits per million tokens = 1,000,000 / tokens per credit. Dollars per million = credits per million x price per credit. A step's credits = input tokens / input rate + output tokens / output rate + 1 operation credit, before any rounding Make applies.
What does each AI model cost per million tokens in Make?
At $0.0009 a credit, a million output tokens costs $0.40 on GPT-5 nano, $10 on Claude Sonnet 5, $25 on Claude Opus 5 and $50 on Claude Fable 5. Output tokens cost 4.5 to 8 times more than input tokens on every row.
| Model in Make's AI Provider | Input tokens per credit | Output tokens per credit | Credits per 1M input | Credits per 1M output | $ per 1M input / output at $0.0009 |
|---|---|---|---|---|---|
| Small tier (GPT-5 nano, minimal reasoning) | 18,080 | 2,260 | 55 | 442 | $0.05 / $0.40 |
| Medium tier (GPT-5 nano, low reasoning) | 18,080 | 2,260 | 55 | 442 | $0.05 / $0.40 |
| Large tier (GPT-5 mini) | 3,616 | 452 | 277 | 2,212 | $0.25 / $1.99 |
| OpenAI GPT-5.6 Sol | 180 | 30 | 5,556 | 33,333 | $5.00 / $30.00 |
| OpenAI GPT-5.6 Sol, over 272,000-token context | 90 | 20 | 11,111 | 50,000 | $10.00 / $45.00 |
| OpenAI GPT-5.6 Terra | 452 | 75 | 2,212 | 13,333 | $1.99 / $12.00 |
| OpenAI GPT-5.6 Terra, over 272,000 | 226 | 50 | 4,425 | 20,000 | $3.98 / $18.00 |
| OpenAI GPT-5.6 Luna | 4,520 | 753 | 221 | 1,328 | $0.20 / $1.20 |
| OpenAI GPT-5.6 Luna, over 272,000 | 2,260 | 502 | 442 | 1,992 | $0.40 / $1.79 |
| OpenAI GPT-5.5 | 180 | 30 | 5,556 | 33,333 | $5.00 / $30.00 |
| OpenAI GPT-5.5, over 272,000 | 90 | 20 | 11,111 | 50,000 | $10.00 / $45.00 |
| OpenAI GPT-5.4 nano | 4,520 | 723 | 221 | 1,383 | $0.20 / $1.24 |
| OpenAI GPT-5.4 mini | 1,205 | 200 | 830 | 5,000 | $0.75 / $4.50 |
| OpenAI GPT-5.4, up to 272,000 | 361 | 60 | 2,770 | 16,667 | $2.49 / $15.00 |
| OpenAI GPT-5.4, over 272,000 | 180 | 40 | 5,556 | 25,000 | $5.00 / $22.50 |
| OpenAI GPT-5.2 | 516 | 64 | 1,938 | 15,625 | $1.74 / $14.06 |
| OpenAI GPT-5.1 | 723 | 90 | 1,383 | 11,111 | $1.24 / $10.00 |
| OpenAI GPT-5 | 723 | 90 | 1,383 | 11,111 | $1.24 / $10.00 |
| OpenAI GPT-5 mini | 3,616 | 452 | 277 | 2,212 | $0.25 / $1.99 |
| OpenAI GPT-5 nano | 18,080 | 2,260 | 55 | 442 | $0.05 / $0.40 |
| Anthropic Claude Fable 5 | 90 | 18 | 11,111 | 55,556 | $10.00 / $50.00 |
| Anthropic Claude Sonnet 5 | 452 | 90 | 2,212 | 11,111 | $1.99 / $10.00 |
| Anthropic Claude Opus 5 | 180 | 36 | 5,556 | 27,778 | $5.00 / $25.00 |
| Anthropic Claude Opus 4.8 | 180 | 36 | 5,556 | 27,778 | $5.00 / $25.00 |
| Anthropic Claude Opus 4.7 | 180 | 36 | 5,556 | 27,778 | $5.00 / $25.00 |
| Anthropic Claude Sonnet 4.6 | 301 | 60 | 3,322 | 16,667 | $2.99 / $15.00 |
| Anthropic Claude Opus 4.6 | 181 | 36 | 5,525 | 27,778 | $4.97 / $25.00 |
| Anthropic Claude Opus 4.5 | 181 | 36 | 5,525 | 27,778 | $4.97 / $25.00 |
| Anthropic Claude Haiku 4.5 | 904 | 181 | 1,106 | 5,525 | $1.00 / $4.97 |
| Anthropic Claude Sonnet 4.5 | 301 | 60 | 3,322 | 16,667 | $2.99 / $15.00 |
Two patterns stand out. First, the cheapest and dearest rows are 125 times apart on output (442 credits per million on GPT-5 nano against 55,556 on Claude Fable 5). Second, contexts over 272,000 tokens cost about double on input for the OpenAI models that list them. Make says these rates "are subject to change as newer models become available", and it already changed the tier rates once this year: on August 25, 2026 its token pricing update moved the Small tier from 5,000 total tokens per credit to 18,080 input and 2,260 output.
How does Make's AI price compare with buying Claude tokens directly?
At $0.0009 a credit, all ten Claude rows match Anthropic's list price to within 1%. Claude Sonnet 5 works out to $1.99 input and $10.00 output per million tokens in Make, against $2 and $10 from Anthropic.
Claude output price per million tokens
| Claude model | Make at $0.0009 a credit (input / output per 1M) | Anthropic list price (input / output per 1M) |
|---|---|---|
| Claude Fable 5 | $10.00 / $50.00 | $10 / $50 |
| Claude Opus 5, 4.8 and 4.7 | $5.00 / $25.00 | $5 / $25 |
| Claude Opus 4.6 and 4.5 | $4.97 / $25.00 | $5 / $25 |
| Claude Sonnet 4.6 and 4.5 | $2.99 / $15.00 | $3 / $15 |
| Claude Sonnet 5 | $1.99 / $10.00 | $2 / $10 |
| Claude Haiku 4.5 | $1.00 / $4.97 | $1 / $5 |
We did not run the same check on the OpenAI rows. OpenAI's pricing page lists promotional and context-dependent prices for some of these models, and we could not match every Make row to one clear list price, so we leave that comparison out instead of guessing.
What does your Make plan's credit price change?
Because Make prices Claude tokens at list price when a credit costs $0.0009, a cheaper credit means cheaper tokens. On Core at $0.000759 a credit you pay about 16% below list; on extra credits bought with the $9 plan you pay 25% above it.
Claude Sonnet 5 output price by credit price
| How you pay for the credit | Price per credit | Claude Sonnet 5, $ per 1M output tokens | Against Anthropic's $10 list price |
|---|---|---|---|
| Extra credits on the $9 plan (+25%) | $0.001125 | 11,111 x $0.001125 = $12.50 | 25% more |
| $9 plan, 10,000 credits | $0.0009 | 11,111 x $0.0009 = $10.00 | Same |
| Core, 150,000 credits for $113.85 | $0.000759 | 11,111 x $0.000759 = $8.43 | About 16% less |
| 500,000 credits for $338.13 | $0.00067626 | 11,111 x $0.00067626 = $7.51 | About 25% less |
There is a catch on top. Every AI step through Make's AI Provider also costs one credit for the operation, and Make's table has no lower rate for cached input. Anthropic charges cache hits at a tenth of the base input price or less, so a long system prompt reused on every call can cost less direct. Make did add prompt caching to its AI Agent app on October 9, 2026; our news report on Make AI Agent prompt caching covers what it does.
What does a typical AI step cost on each model?
For an illustrative step of 2,000 input tokens and 300 output tokens, credits range from 1.2 on GPT-5 nano to 39.9 on Claude Fable 5. At the Core price, that is $0.0009 to $0.0303 per step.
Credits for one AI step by model
| Model | Credits per step | Cost per step at Core | Cost of 1,000 steps at Core |
|---|---|---|---|
| GPT-5 nano (also Small and Medium tiers) | 2,000 / 18,080 + 300 / 2,260 + 1 = 1.24 | $0.0009 | $0.94 |
| GPT-5.6 Luna | 1.84 | $0.0014 | $1.40 |
| GPT-5 mini (Large tier) | 2.22 | $0.0017 | $1.68 |
| GPT-5.4 mini | 4.16 | $0.0032 | $3.16 |
| Claude Haiku 4.5 | 4.87 | $0.0037 | $3.70 |
| Claude Sonnet 5 | 8.76 | $0.0066 | $6.65 |
| Claude Sonnet 4.6 | 12.64 | $0.0096 | $9.60 |
| Claude Opus 5 | 20.44 | $0.0155 | $15.52 |
| GPT-5.6 Sol | 22.11 | $0.0168 | $16.78 |
| Claude Fable 5 | 39.89 | $0.0303 | $30.28 |
Credits matter for more than money. On a 10,000-credit plan, 1,000 steps on Claude Fable 5 would use about 40,000 credits, four months of the plan, and Make stops every scenario when credits run out. Our FAQ on what happens when Make runs out of credits explains what stops and how to restart it.
Should you use Make's AI Provider or your own API key?
Use your own key when your credits cost $0.0009 or more, when you rely on cached prompts, or when volume is high. Make's AI Provider is simpler and, on cheaper credit tiers, priced below Claude's list price. Free plans can only use Make's provider.
Make's credits guide sets the two billing routes. With Make's AI Provider, you pay Make credits for the operation and the tokens. With a custom AI provider connection, which Make made available on all paid plans in November 2025, you pay Make one credit per operation and pay the AI provider for tokens directly.
| Situation | Cheaper route, from the numbers above |
|---|---|
| Free plan | Only Make's AI Provider is available |
| Running on extra credits most months | Your own key, on small plans: extra credits on the $9 plan put tokens 25% above list (on Core extras, $0.000949 a credit, about 5% above) |
| Core or higher tier, light AI use, no caching | Make's AI Provider: tokens about 16% to 25% below Claude's list price |
| Long fixed prompts reused on every call | Your own key with the provider's prompt caching, if it is cheaper for your prompt mix |
| You need one bill and no provider account | Make's AI Provider |
A third route is to move AI-heavy steps out of Make. Custom AI agent development or an n8n workflow calling the model directly can make sense once AI steps dominate the bill. Our Make vs n8n comparison covers how the two platforms bill for AI agents.
What does this data prove, and what does it not?
It proves what Make charges per token, by its own published table on October 30, 2026, and that Claude tokens match Anthropic's list price at $0.0009 a credit. It does not show real token counts, output quality or what your plan costs.
- It does show the relative cost of every model Make lists, the size of the gap between cheap and expensive models, and that the credit price you pay sets your token price.
- It does not show how many tokens your own steps use. Reasoning models can use many more output tokens than the 300 we assumed.
- It does not compare quality. A cheaper model that gets answers wrong costs more in rework than it saves.
- It does not cover knowledge files, embedding tokens, Make AI Web Search or Content Extractor, which Make bills on other rules, or Maia, whose credits per message vary by task.
- Plan prices are Make's own examples, taken from its Help Center. Your tier and currency may differ; check the live pricing page.
More questions about Make AI credit costs
How many tokens does one Make credit buy?
It depends on the model and on whether the token is input or output. On Make's Small and Medium tiers, one credit buys 18,080 input tokens or 2,260 output tokens. On Claude Sonnet 5 it buys 452 input or 90 output. On Claude Fable 5 it buys 90 input or 18 output. Make lists every rate in its credits guide, last updated October 29, 2026.
Does Make charge extra for AI on top of tokens?
Yes, one credit per operation. Make's guide to how features use credits says Make AI Agents and Make AI Toolkit use one credit per operation plus credits for tokens when you use Make's AI Provider. With your own provider connection on a paid plan, you pay one credit per operation and the provider bills the tokens. Knowledge files add embedding tokens either way.
Is Make's AI Provider more expensive than OpenAI or Anthropic directly?
For Claude models it depends on your credit price. At $0.0009 a credit, Make matches Anthropic's list price. Below that, as on Core at $0.000759, Make's tokens cost less than list; on extra credits bought with the $9 plan they cost 25% more. Direct API features such as prompt caching discounts can still make your own key cheaper. We did not verify the OpenAI rows.
Which AI model should a Make scenario use?
Start with the cheapest model that passes your test cases, then move up only where answers fail. Sorting, tagging and short extraction often work on small models, which cost 1.2 to 2.2 credits a step in our example. Keep large models for steps that need judgement. Our guide to testing an AI agent before launch shows how we build those test cases.
Can you set up AI steps in our Make account?
Yes. The strategy call and audit are free, and we check which model and billing route fits each step before we build. A small build of two or three scenarios is about $1,000 at the time of writing, and larger systems get a fixed price after the audit. Everything runs on your own Make account and your own AI keys.
Sources and further reading
- Make Help Center, Credits (Make's AI Provider input and output tokens per credit for 3 tiers and 27 models; custom connections on paid plans; updated October 29, 2026), read October 30, 2026.
- Make Help Center, How features use credits (1 credit per operation plus token credits for Make AI Agents and AI Toolkit), read October 30, 2026.
- Make release note, Updated Make's AI Provider token pricing model (separate input and output rates from August 25, 2026), read October 30, 2026.
- Make Help Center, Extra credits ($9 plan with 10,000 credits example; 25% additional cost), read October 30, 2026.
- Make Help Center, Change your subscription (Core $113.85 for 150,000 credits example), read October 30, 2026.
- Make release note, Adjustments to plans and pricing ($338.13 for 500,000 credits; custom AI connections on all paid plans from November 6, 2025), read October 30, 2026.
- Anthropic, Claude API pricing (base input and output prices per million tokens; cache hit pricing), read October 30, 2026.
- Make Help Center, Credit usage for AI agents (knowledge files, embedding tokens and file description tokens), read October 30, 2026.
Make says its models and token rates change as newer models arrive. Recheck the credits guide before you budget, and treat this table as a snapshot dated October 30, 2026.
