Last updated: November 20, 2026
Usage-based pricing bills customers for what they actually use, like requests, messages, tokens or tasks, rather than a flat fee per user. AWS, Twilio and AI model providers all charge this way, and it’s moving quickly into everyday SaaS through credits. Below are real prices, the math behind them, and a way to decide whether the model suits your product.
Short answer: With usage-based pricing, the bill goes up and down with consumption: per request, per message, per token or per task. It works best when customers get more value the more they use, and when you can measure that use reliably. Most SaaS companies now pair it with a base fee or credits instead of going fully pay-as-you-go.
Key takeaways
- After a free tier of 1 million requests a month, AWS Lambda charges $0.20 per 1 million requests.
- In the U.S., Twilio charges $0.0083 per SMS segment, plus carrier fees of up to $0.005 per message.
- Through the API, Claude Sonnet 5.5 runs $2 per million input tokens and $10 per million output tokens.
- In early 2026, 79 of the 500 SaaS and AI companies PricingSaaS tracks offered credits, up from 35 at the end of 2024.
- A usage metric only works if customers can predict their bill, so allowances, alerts and caps matter as much as the price.
Table of Contents
What usage-based pricing is
Usage-based pricing, sometimes called consumption pricing or pay-as-you-go, links the price to a unit of consumption. You choose a value metric, price each unit, and bill for the units used each period.
The unit depends on the product:
- Infrastructure: requests, compute time, storage or data transfer.
- Communications: messages, minutes or emails sent.
- AI models: tokens of input and output.
- Automation tools: tasks, runs or workflow steps.
- Business software: credits that customers spend on features, AI features in particular.
Billing tends to work one of two ways. In arrears, the vendor meters usage through the month and invoices afterward, the way most cloud and messaging bills work. Prepaid, the customer buys credits or a commitment upfront and spends it down, which makes things more predictable for both sides. Lots of hybrid plans do both: a prepaid monthly allowance, then billing in arrears for anything over it.
It’s the opposite of seat-based pricing, where a customer pays per named user per month however much each person uses the product. Our look at Jira’s 2026 price increase shows seat-based pricing in action: the bill is easy to forecast, but it grows with headcount, not value.
Usage-based pricing vs seat-based pricing and other models
Software gets priced five common ways today, and most real pricing pages combine two of them.
| Model | How you charge | Best for | Main risk |
|---|---|---|---|
| Seat-based pricing | A fixed fee per user per month | Collaboration tools where every user gets similar value | Revenue stalls when customers cut headcount or share logins |
| Usage-based pricing | A price per unit consumed | APIs, infrastructure, messaging, AI models | Bills are hard to forecast, for you and for the customer |
| Hybrid pricing | A platform or seat fee plus usage above an allowance | SaaS products adding AI or automation features | Two things to explain on one invoice |
| Credits | Customers buy or receive credits and spend them on actions | Products with many different usage types | Credit values feel abstract unless you translate them into tasks |
| Outcome-based | A fee per result, such as a resolved ticket | Products that own the whole job end to end | Hard to agree on what counts as an outcome |
Seat-based pricing isn’t disappearing. It still suits tools where value follows people. Once software starts doing work by itself, through automation or AI, seat counts stop reflecting the value delivered, and that’s the opening for usage-based pricing.
7 real usage-based pricing examples (as of November 2026)
The quickest way to get the model is to look at real price points and do the math. The first three examples are pure usage pricing. The last four are SaaS companies shifting toward hybrid pricing and credits.
| Company | What is metered | Price checked November 4, 2026 | Free allowance |
|---|---|---|---|
| AWS Lambda | Requests and compute time | $0.20 per 1 million requests; $0.0000166667 per GB-second (x86, US East) | 1 million requests and 400,000 GB-seconds per month |
| Twilio SMS (U.S.) | Message segments | $0.0083 per segment, plus carrier fees of $0.0035 to $0.005 | None listed |
| Claude Sonnet 5.5 API | Input and output tokens | $2 per million input tokens; $10 per million output tokens | Batch API discount of 50% |
AWS Lambda. A function that runs 3 million times a month, 200 milliseconds each, with 512 MB of memory, uses 300,000 GB-seconds. That sits inside the free allowance, so compute is free and you pay $0.40 for the 2 million requests over the free tier. Push the same function to 30 million runs and the bill climbs to about $49.13: $43.33 for compute and $5.80 for requests. That’s ten times the usage for roughly 120 times the bill, because the free tier no longer soaks it up.
Twilio. Texting 10,000 AT&T customers in the U.S. costs 10,000 x ($0.0083 + $0.0035) = $118 before volume discounts. Twilio says volume and committed-use discounts exist but sends you to sales for the figures, and it charges $0.001 for messages that end up in a failed status.
Claude API. A job that sends 10 million input tokens and gets 2 million output tokens back costs 10 x $2 + 2 x $10 = $40, or $20 through the Batch API. Output tokens cost five times as much as input tokens, which is why long answers, not long prompts, drive most AI bills.

Hybrid pricing and credits: where SaaS moved in 2025
Pure usage pricing is common in infrastructure. In mainstream SaaS the move has mostly been toward hybrid pricing and credits.
A January 2026 analysis by PricingSaaS in Growth Unhinged found that the top 500 SaaS and AI companies with public pricing made more than 1,800 pricing changes in 2025, roughly 3.6 each. 43% were packaging changes, 40% changed prices or price structure, and 17% changed usage limits. Credit models grew fastest: 79 of the 500 offered credits at the time of writing, up from 35 at the end of 2024.
The same analysis lists four moves worth knowing:
- Notion rolled its $8 per user AI add-on into the Business plan and lifted Business from $15 to $20 per user per month.
- Slack folded its $10 per user AI add-on into its core plans, with different AI features at each tier.
- Atlassian rolled its $20 per month Rovo add-on into its products, with credit limits it wasn’t yet enforcing.
- Salesforce launched AI credits in May 2025.
The pattern is easy to spot. Companies hold on to a predictable base price, then use credits to meter the costly part, usually AI. That protects margins without putting customers on a bill that changes every month.
A November 2026 example: Atlassian’s Rovo credits
Atlassian’s Jira plans show the hybrid model working. Every paid Jira plan comes with a monthly pool of Rovo AI credits: 25 per user on Standard, 70 on Premium and 150 on Enterprise, per the Jira pricing page we checked on November 7, 2026. Credits are pooled across the organization and reset monthly. Rovo Search doesn’t touch them, while Chat and Agent requests do. Atlassian says it isn’t charging for usage beyond the allowance today and has promised at least 90 days’ notice and an explicit opt-in before that changes.
If you’re moving toward usage-based pricing, that pattern is worth borrowing: seats for predictability, a usage allowance for AI features, and a clear path with advance notice before any overage charge. Our Jira pricing guide breaks down the full plan prices.
How to choose a usage metric customers will accept
Choosing the metric is the biggest decision in usage-based pricing. Get it wrong and you’ll spend years explaining invoices. A good metric passes five tests:
- It tracks value. When the customer gets more done, usage rises. Messages delivered pass. Logins don’t.
- The customer can predict it. A finance team should be able to estimate next quarter’s usage from its own plans.
- You can measure it accurately. If you can’t count it in real time and audit it later, you can’t bill for it.
- It’s hard to game. Customers shouldn’t be able to lower the bill by changing behavior that doesn’t affect the value they get.
- It scales with your cost. For AI and infrastructure products, the metric needs to rise when your costs rise, or gross margins vanish at scale.
A fast way to test options is to write down three or four possible metrics and score each against the five tests. Tokens do well on cost but badly on predictability. Resolved tickets do well on value but badly on measurement, because someone has to agree what “resolved” means. Most teams land on a hybrid: a base plan for predictability and a metered unit for the part that varies. Once it’s live, watch expansion revenue and churn in your core SaaS metrics, because that’s where a pricing change shows up.
A worked example: when usage beats seats
Here’s a simple illustration with invented numbers, not any vendor’s actual prices.
Picture a support automation tool sold to a 20-agent team. Option one is seat-based pricing at $50 per agent per month, so $1,000 a month. Option two is usage-based pricing at $0.50 for each ticket the tool resolves by itself.
- At 1,500 resolved tickets a month, usage pricing comes to $750, saving the customer $250.
- At 2,000 tickets, both options cost $1,000. That’s break-even.
- At 3,000 tickets, usage pricing comes to $1,500, and the vendor earns more because it delivered more.
Now add a cap. If the contract limits usage charges to $1,200 a month, the customer never pays more than 20% over the seat price, and the vendor still earns more in busy months. A cap like that is often what gets a usage deal past procurement.
That’s why vendors like usage-based pricing for automation. As the product does more of the work, the customer may need fewer seats, so seat revenue shrinks right when the product is succeeding. Pricing on the outcome keeps revenue tied to value. The customer still needs a cap or a commitment, though, or a busy month turns into a budget shock.

How to roll out usage-based pricing without scaring customers
The price is only half the job. The other half is making the bill feel safe, and five practices keep showing up on the pricing pages above:
- Give a free or included allowance. AWS includes 1 million Lambda requests a month, so customers can try the product before the meter matters.
- Show usage in the product. Customers should be able to see current usage and a month-end forecast without opening an invoice.
- Offer alerts and hard caps. Let the customer set a budget, get a warning at 80% and decide whether to stop or carry on.
- Sell commitments at a discount. Twilio offers committed-use discounts. Prepaid commitments give the customer a lower rate and give you predictable revenue.
- Translate units into outcomes. “1,000 credits” doesn’t mean much. “About 200 AI summaries” does.
If you’re moving existing customers off seat-based pricing, run both models against last year’s usage data first. Any account whose bill would jump needs a conversation before the change, not after.
Mistakes to avoid with usage-based pricing
- Picking a metric customers can’t control. When usage depends on things outside the customer’s hands, every invoice becomes an argument.
- No ceiling. Unlimited exposure frightens procurement teams, and a cap or commitment usually costs less than a lost deal.
- Hiding the unit price. Customers compare per-unit prices. If yours is buried, they’ll assume it’s high.
- Ignoring your own costs. AI products charging a flat fee for unlimited use can lose money on their heaviest users.
- Changing the metric too often. The PricingSaaS data shows companies making several pricing changes a year. Tweaking limits is fine. Changing what you meter confuses customers and wrecks their budgets.
For buyers: keeping a usage-based bill under control
Usage-based pricing moves risk onto the buyer, so finance and operations teams should handle these contracts differently from seat licenses.
- Model three scenarios before you sign: expected usage, a busy month and the worst case.
- Get caps and alerts in writing. Our AI vendor questionnaire has pricing and usage questions you can send before buying.
- Watch the expensive unit. With AI tools, that’s usually output tokens or premium actions.
- Only negotiate commitments after a few months of real data. A commitment built on guesses usually ends in unused credits.
- Review usage monthly. Usage bills drift. Fifteen minutes a month catches a runaway workflow before it eats a quarter’s budget.

Our Take / Expert Perspective
Our take: it’s no longer seats versus usage. The 2025 data points to a settled middle ground where companies keep a predictable base price and meter the part of the product whose cost grows with use, which today mostly means AI. Credits are the tool that makes it work.
Who wins? Vendors whose products do real work on their own, because usage pricing lets revenue grow with the work instead of with headcount. Who loses? Buyers who sign usage contracts with no caps or alerts, and vendors charging a flat fee for AI features whose costs climb with every request.
We expect two things over the next 12 months, as a forecast rather than a certainty. First, more SaaS companies will fold AI add-ons into core plans with credit allowances, as Notion, Slack, Airtable and Atlassian have done. Second, enforcement will tighten. Atlassian wasn’t enforcing its credit limits when PricingSaaS published its analysis, and “not yet enforced” rarely lasts once costs hit the margins.
What to do: if you sell software, find the one unit that follows both customer value and your own cost, and wrap it in an allowance, an alert and a cap. If you buy software, ask every vendor with credits what one credit gets you today, and get the answer in writing.
Frequently Asked Questions
What is an example of usage-based pricing?
AWS Lambda is a clean example. After a free monthly allowance of 1 million requests and 400,000 GB-seconds, it charges $0.20 per 1 million requests plus a fee for compute time. Twilio’s per-message SMS pricing and per-token AI model pricing follow the same logic.
Is usage based pricing better than seat-based pricing?
Neither wins across the board. Usage pricing suits products whose value grows with consumption, like APIs, messaging and AI. Seat-based pricing suits tools where every user gets similar value and customers want a predictable bill.
What is hybrid pricing?
Hybrid pricing pairs a fixed fee, such as a platform or per-seat fee, with usage charges above an included allowance. Customers get a predictable base, and the vendor can still charge for heavy use.
How do credits work in SaaS pricing?
Customers get or buy credits and spend them on actions, often AI features, so one plan can cover lots of usage types. The PricingSaaS analysis found 79 of 500 SaaS and AI companies using credit models in early 2026, up from 35 at the end of 2024.
How can a customer control a usage-based bill?
Set budgets and alerts, push for hard caps and review usage monthly. Once you’ve got a few months of real data, negotiate a commitment for a lower unit price.
What is usage-based billing?
Usage-based billing is the system that measures what each customer used in a billing period and turns it into an invoice at the agreed rates. The two terms are often used interchangeably, but usage-based pricing sets the rates, such as a price per API call or per gigabyte, while billing does the metering, rating and invoicing. You need both a clear price list and accurate metering before you switch.
Sources
- Amazon Web Services: AWS Lambda pricing, checked November 4, 2026
- Twilio: SMS pricing in the United States, checked November 4, 2026
- Anthropic: Claude Sonnet 5.5 model overview, checked November 4, 2026
- Growth Unhinged and PricingSaaS: The 2025 state of SaaS pricing changes, published January 7, 2026, checked November 4, 2026
Photos: Jon Moore, Jakub Żerdzicki, Shoper and Jason Briscoe on Unsplash.







Leave a Reply