Make's move to credits is a rename, not a price rise. Your AI steps are the real bill.

Make now uses two words where it used one, and it uses them for different jobs. Credits are what you buy and spend. Operations are still there, but they now describe the outcome of your activity on the platform rather than the thing you are billed for. Make's own framing is that credits help you track what you are spending, and operations show what happened.

That sounds like semantics until you notice why the split was needed. A single word could not carry both meanings once AI modules arrived, because an AI module performs one operation and can consume thirty credits doing it. Splitting the vocabulary was the precondition for metering AI, and metering AI is the actual event here.

The pattern is not unique to Make. Notion, ClickUp and Zapier all moved their AI features onto credit or token style meters inside about a year, and we have written up the same shift in execution based versus task based automation pricing. Every vendor has landed in the same place: the deterministic part of the product stays priced per action, and the AI part gets priced per unit of consumption. Make just did the vocabulary cleanup more visibly than most.

That is the whole story in one table. One ordinary non-AI module costs 1 credit. The identical AI step costs a fifth of a credit or thirty credits depending on a dropdown nobody revisits, and the difference between the cheapest and dearest row is 150x. Not 15 percent. 150 times.

Two things fall out of it that are worth more than the headline. The first is that GPT-5 mini lands at almost exactly 1 credit per run, the same as a plain module. If you have been avoiding AI steps because you could not forecast them, that row is your answer: pick the Large tier model, and an AI step costs what a Google Sheets row costs.

The second is that output tokens dominate. Every model on the list bills output at roughly a fifth to a sixth of the input rate, so a chatty prompt is cheap and a chatty answer is not. In the Sonnet 5 row, 300 output tokens cost 5 credits while 1,200 input tokens cost 4. Telling the model to reply in one sentence rather than one paragraph is a bigger lever on your Make bill than anything you will do to the prompt.

None of this is an argument for always picking the small model. A ticket triage that misclassifies is worse than useless, and if Sonnet 5 gets it right and nano does not, 9 credits is cheap. The argument is that the choice should be deliberate and priced, which is exactly what the credit rename now lets you do. Under operations you could not see this at all.

Do not migrate. If a rename triggered a platform review, the review was already overdue for other reasons. Make against n8n against Zapier is a real decision with real trade-offs, and we have written it up properly in Zapier Agents vs Make AI Agents vs n8n and in what each one does when a workflow fails. A vocabulary change on an invoice is not on that list.

Do go and read your own tags. Open the scenarios that run most often and look for the dynamic usage labels. You are looking for tokens, file size, per page and run time. That is your entire variable cost, and it is usually three or four modules out of several hundred.

Do make the model an explicit decision. For each of those AI steps, ask what the job actually needs. Classification, extraction and routing generally do not need a frontier model, and the table above shows what that choice is worth. Drafting that a customer will read usually does. Both answers are fine. Defaulting is not.

Do cap your outputs, because output tokens carry most of the cost on every model Make lists.

Consider bringing your own key if predictability matters more than a single invoice. Third-party AI connections cost 1 credit per operation on Make and move the token cost to your provider, which is the right shape when finance wants automation spend and AI spend separated.