Affiliate disclosure: This post contains affiliate links. If you sign up for Make through Make, Automated Wealth Lab may earn a commission at no extra cost to you. Prices below were checked against published sources in October 2026.
Make changed how its built-in AI Provider burns credits — and most summaries of the change are wrong about who wins. The common take (“AI got cheaper”) is half the story. The real story: Make stopped pricing AI as one blended token rate and split it into separate input and output token rates. Feed a model a long document and ask for a short answer, and your cost dropped dramatically. Ask for a long essay back, and your cost went up.
The change took effect August 25, 2026 for Make’s AI Provider Small, Medium, and Large models (confirmed in Make’s own help article; it hit most users’ September bills, which is why it reads as a “September 2026” change). If you bring your own API key, nothing changed — this doesn’t affect apps where you use your own key.
Let’s do the actual math.
Old vs new rates: the full table
Previously, Make priced the Small, Medium, and Large tiers on total tokens (input + output combined). Now input and output tokens get their own rates — the same approach Make already used for its other AI models.
| Model tier | Old: combined tokens / credit | New: input tokens / credit | New: output tokens / credit |
|---|---|---|---|
| Small | 5,000 | 18,080 | 2,260 |
| Medium | 3,500 | 18,080 | 2,260 |
| Large | 1,500 | 3,616 | 452 |
The headline numbers, per tier:
- Input tokens got roughly 3–5x cheaper. Small went from 5,000 to 18,080 tokens per credit — a 72.4% cost cut per input token. Medium went from 3,500 to 18,080 (an 80.6% cut). Large went from 1,500 to 3,616 (a 58.5% cut).
- Output tokens got more expensive. Small went from 5,000 to 2,260 tokens per credit — output now costs about 2.2x what it did before. Medium: 3,500 → 2,260 (1.55x). Large: 1,500 → 452 (3.3x).
There is no single “cheaper or more expensive” verdict. It depends entirely on the ratio between what you send and what the model sends back.
The 73% figure, explained honestly
Where does the “73% cheaper” claim come from? It’s the Small and Medium tier input-token improvement rounded: a workflow that is almost entirely input — feeding huge context to the model and getting tiny answers back — pays ~72–73% less for that input portion than it did before.
Two caveats that keep this honest:
- It’s the input side only. A real workflow has both. A 73% cheaper input token doesn’t mean a 73% cheaper scenario — unless your scenario is nearly all input.
- Large tier gets less. At 3,616 vs 1,500, Large’s input improvement is ~59%, not 73%.
So the claim holds as: input-heavy workflows on Small/Medium got up to ~73% cheaper on their dominant cost line. Anything beyond that is marketing.
Three workflow archetypes, repriced
Here are three common solo-founder scenarios, repriced with the published rates. Math uses the Small tier (the default most automations reach for).
1. The Extractor (the big winner)
Scenario: Watch a Gmail inbox for new invoices, send each email body to the AI module to extract vendor, date, total, and line items into structured fields, then write to a spreadsheet.
- Tokens per run: ~4,000 input (email body + extraction instructions), ~150 output (a few JSON fields).
- Old cost: 4,150 ÷ 5,000 = 0.83 credits/run.
- New cost: 4,000 ÷ 18,080 + 150 ÷ 2,260 = 0.29 credits/run.
- Result: ~65% cheaper. At 1,000 invoices a month, that’s ~540 fewer AI credits.
This is the archetype the repricing was aimed at: summarize, extract, classify, tag, score, route. High-input, low-output. If your automation reads documents and answers in a handful of fields, you won.
2. The Generator (the loser)
Scenario: Take a one-line product note and generate a full 800-word SEO product description.
- Tokens per run: ~300 input (short prompt), ~1,100 output (the description).
- Old cost: 1,400 ÷ 5,000 = 0.28 credits/run.
- New cost: 300 ÷ 18,080 + 1,100 ÷ 2,260 = 0.017 + 0.49 = 0.51 credits/run.
- Result: ~82% more expensive.
Any workflow whose value is long generated text — blog drafts, product descriptions, long-form replies — now costs meaningfully more on Make’s built-in models. This is the scenario nobody warned you about.
3. The Mixed Agent (roughly neutral)
Scenario: A support-triage agent that reads a ticket (~1,500 input tokens) and writes a draft reply (~500 output tokens).
- Tokens per run: 1,500 input, 500 output.
- Old cost: 2,000 ÷ 5,000 = 0.40 credits/run.
- New cost: 1,500 ÷ 18,080 + 500 ÷ 2,260 = 0.083 + 0.221 = 0.30 credits/run.
- Result: ~24% cheaper. Balanced ratios get a mild discount; the cheaper input outweighs the pricier output.
The audit checklist: find AI-credit waste
This repricing changed which module settings are wasteful. Run this checklist on your scenarios:
- Check which model tier each AI module uses. Small and Medium now share identical rates (18,080 in / 2,260 out). If a module sits on Large for a simple extraction, downgrade it: Large’s input rate (3,616) is now 5x worse than Small’s (18,080).
- Trim system prompts, not just user prompts. System instructions burn input tokens every run. Prompts you can cut in half directly cut your dominant cost line for extraction workflows.
- Cap output length where output is the product. If your scenario generates text, add explicit max-length instructions (“answer in under 100 words”). Output is now the expensive token.
- Reconsider “summarize then generate” chains. The first module just got much cheaper; the second got more expensive. Re-run the math per module instead of assuming.
- Compare against your own API key. If your generation-heavy scenario got pricier, a direct API call via the HTTP module may now undercut Make’s built-in provider for long-output work. Price it both ways.
What didn’t change
- Non-AI module operations still cost one credit each (Router and Filter excluded). The AI repricing touches only the AI Provider’s Small/Medium/Large tiers.
- Plan pricing didn’t change. As of October 2026: Free at $0 (1,000 credits), Core from $9/mo (10,000 credits, annual billing), Pro from $16/mo, Teams from $29/mo.
- Bring-your-own-key scenarios are untouched. If you call an external LLM with your own API key, this whole article doesn’t apply to you.
FAQ
Q: Did Make’s AI actually get 73% cheaper?
A: Input tokens on the Small and Medium tiers got ~72–80% cheaper per token. Whether your scenario got cheaper depends on your input-to-output ratio. “Up to ~73% cheaper for input-heavy workflows” is the honest version.
Q: Which of my scenarios should I check first?
A: Sort by AI module usage, then by model tier. Anything on Large doing simple classification or extraction is the biggest waste — move it to Small/Medium.
Q: Does this affect Make’s other AI models or AI Agents?
A: The published change covers the AI Provider’s Small, Medium, and Large models. Make’s other models already used separate input/output rates.
Q: Should I switch from Make’s built-in AI to my own API key?
A: For extraction (long input, short output), Make’s built-in Small tier at 18,080 input tokens per credit is genuinely competitive. For long-output generation, a direct API call often wins. Price your actual token mix both ways.
Q: When did this take effect?
A: August 25, 2026, per Make’s own documentation. It showed up in most users’ usage and bills starting September.
Methodology
This article is a documented price-sheet analysis, not a hands-on test. Rate changes were taken from Make’s own help article and cross-checked against independent coverage (IBTimes SG; DEV community pricing guides, September–October 2026). Workflow archetype math is arithmetic applied to published rates with stated example token counts — illustrative estimates, not measured from live runs. What was NOT done: no Make scenarios were executed, no credit consumption measured. Affiliate terms from a third-party directory (verified 2026-06-29), not Make’s partner portal.
AWL verdict: If your Make automations read more than they write — invoice extraction, email triage, lead scoring, document summarization — the August 2026 repricing quietly cut your AI credit burn by up to ~73% on the input side. If your automations generate long text, costs went up ~50–80% on those modules — cap output lengths or move generation to your own API key. Either way, audit model tiers first: anything still on Large for simple extraction burns 5x the input credits it needs.
Sources
- https://www.ibtimes.sg/ai-automation-platform-make-changes-credit-rates-which-scenarios-now-cost-more-less-94021
- https://dev.to/stimlau/make-pricing-in-2026-free-core-pro-teams-and-credit-costs-lcm
- https://dev.to/2pizza/make-pricing-explained-what-you-actually-pay-59p3
- https://github.com/lxcardoza993/awesome-oss-sponsorship/blob/HEAD/docs/affiliate-programs.md
