Right Suite logoRight Suite
PricingNovember 22, 2026

How to Price an AI SaaS When Your Model and API Costs Keep Changing (2026)

Price your AI SaaS on the value buyers get, sell in units they count, and keep a cost buffer you control. Your price stays stable while model costs move.


Price your AI SaaS on what the buyer gets, then protect your margin separately. Sell in a unit the buyer counts, like documents or conversations, and build the price from your heavy-user cost with a buffer on top. Cap or meter the expensive actions. When model costs drop you keep the margin, and when they rise your buffer and limits absorb it. Test the price with buyers before launch in RightPrice.

The trap is tying your price to your API bill. If you price at "cost plus 30%," every provider price change becomes a repricing project, and every price cut you pass on trains buyers to expect the next one.

Test your AI pricing in RightPrice to see whether buyers accept the number before your cost model is final.

Who this is for

You run or are launching a SaaS product where each AI action costs you money: calls to a model provider, your own inference, or both. Your costs have moved more than once. You might switch models next quarter. You still need a price buyers can plan around.

If you're still deciding between seats and usage, read usage-based vs per-seat pricing for a SaaS with AI features first.

Why shouldn't your price follow your API costs?

Because your buyer doesn't care what you pay your model provider. They care what the product does for them. A contract review that saves a lawyer an hour is worth the same to that lawyer whichever model ran it.

Pricing on value gives you two things. Your price stays stable, which buyers like. And when model costs fall, that drop becomes your margin instead of a price cut.

How to price when costs are moving

  1. Measure cost per outcome, not per call. Work out what it costs you to deliver one unit the buyer cares about: one summarised call, one processed invoice, one generated report. Include retries and long inputs.
  2. Find your heavy-user cost. Average cost hides the risk. Price from what your top users cost you each month, because they're the ones who can sink your margin.
  3. Add a buffer for cost increases. Assume your cost per outcome could rise for a while: a better model you'll want to switch to, a provider price change, longer inputs as customers use the product more. Size the buffer so a reasonable rise doesn't push heavy users into a loss.
  4. Set the price from value, then check it against step 3. Start with what the outcome is worth to the buyer. If value-based price clears your heavy-user cost plus buffer comfortably, you're done. If it doesn't, the plan needs a limit.
  5. Put limits on the expensive actions. Include a monthly allowance of the costly AI actions in each plan, with a fixed price for extra packs. Name the allowance in the buyer's unit.
  6. Keep the model choice behind the curtain. Don't promise a specific model on your pricing page unless buyers ask for it. That leaves you free to route simple tasks to cheaper models.
  7. Review cost per outcome every month. Track it by plan. If heavy-user margin gets thin, adjust the allowance on new signups before you touch the price.

What a cost buffer looks like

For example, say you sell an AI tool that turns customer support tickets into weekly insight reports.

LineIllustrative number
Your cost per report today$0.80
Reports a typical customer runs per month20
Reports a heavy customer runs per month120
Heavy-user cost per month$96
Buffer if cost per report rose by half$144
What the insights are worth to a support leadSeveral hours of analysis each week
Price you test$199 a month including 100 reports, $1.50 per extra report

At today's costs the heavy user leaves you a healthy margin. If costs rise by half, you still make money on them, and the overage covers anyone past 100. If costs fall, you keep the difference and can make the allowance more generous as a selling point later.

Symptoms that your AI pricing is broken

SymptomCheckFix
Margin swings month to monthIs your price a markup on cost?Re-anchor to value, meter the costly actions
A handful of accounts cost more than they payDo plans include unlimited AI?Add a fair-use allowance per plan
Buyers ask which model you use before buyingIs model choice part of their quality worry?Show output quality, offer a sample, keep model routing flexible
You delay switching to a better model because of costIs your buffer too thin?Raise the price on new plans or reduce the allowance
Buyers hesitate at overage pricingIs the extra pack price clear and capped?Show a monthly cap and alert before overage

How to test it with RightPrice

Your cost model tells you the lowest price you can afford. It doesn't tell you what buyers will pay. Test that side with RightPrice.

Describe the offer the way a buyer sees it: the outcome, the plan, the allowance and the overage. Add your target audience. RightPrice builds personas for that audience, including a competitor-minded buyer and a skeptic, and runs them through a simulated discussion you watch live. The report shows a confidence score, sentiment breakdown, willingness-to-pay analysis, a price assessment and each buyer's objections.

Run two versions: one with a generous allowance and one with a tighter allowance at a lower price. If buyers accept the higher price with the bigger allowance, you get both a better margin and a buffer. A Quick run takes about 10 minutes and uses 1 credit.

Simulated buyers give you a quick read before real sales calls. They're especially useful here because you can test pricing structures you haven't built billing for yet.

FAQ

Should I pass model price cuts on to customers?

Not as a price cut. Pass them on as a bigger allowance or a new feature on the same price. Buyers get more value, you keep a stable price, and you're not training them to wait for discounts.

What if my costs go up sharply?

Change the allowance for new customers first, and give existing customers notice before changing theirs. Your buffer should cover the gap while you adjust. If it can't, the plan was priced too close to cost.

Should I use credits for AI features?

Credits work when your AI actions vary a lot in cost, because you can price each action in credits and adjust the mapping later. Always tell buyers what one credit gets them in plain terms.

How do I know what the outcome is worth to the buyer?

Ask what they do today without you and what it costs them in time or money. How to price a B2B SaaS product with no customers walks through that calculation.

Your next step

Calculate your cost per outcome for your heaviest user this month, then set an allowance that keeps them profitable if that cost rises by half. Test the plan in RightPrice before you change your billing. For more on picking the model itself, see how to choose a SaaS pricing model.