Guides / per tier token budget
Working out how much usage a subscription tier can carry
A subscription price is fixed and the usage behind it isn't, so every tier has a call count above which it loses money. Working it out takes one division, and skipping it explains how unlimited plans get priced. Culpa, a local-first LLM cost, margin, and forecast ledger, prices each call and attributes it to the customer and plan behind it.
Why this happens
Subscription pricing assumes cost per customer is roughly stable, which was true when the marginal cost of a user was storage. With model spend it isn't: two customers on the same plan can differ by an order of magnitude in what they cost you, and nothing about the plan constrains that. The tier therefore has an implied budget whether or not anyone has written it down, which is the price times your target gross margin, and above it the customer is unprofitable. The reason this bites later rather than sooner is that the average stays comfortable long after the tail stops being. Most customers use a fraction of what they could, so blended margin looks fine while a small group loses money on every call. That group is usually your most engaged users, which makes the fix a product decision rather than a billing one, and it's much easier to make before the plan is public.
What this usually looks like
- A tier is advertised as unlimited and nobody has worked out what it can afford.
- You know blended gross margin and not margin per customer.
- Your heaviest users are on your cheapest plan.
- Nobody can say which customers are unprofitable this month.
- A tier's price was set before the model spend behind it was known.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Mistakes that cost the most
| Mistake | Why it hurts | Do instead |
|---|---|---|
| Pricing a tier from average usage. | The average sits inside a skewed distribution, and the tail is what loses money. | Price from the distribution, and know what your top decile costs on each plan. |
| Advertising unlimited without a fair-use position. | Fixed revenue against unbounded cost has no ceiling, and one account can absorb a tier's margin. | Decide the limit before anyone tests it, even if it stays generous and undisclosed. |
| Reporting blended gross margin only. | It stays healthy while a subset of customers is loss-making on every call. | Report margin per customer, and look at the bottom of the distribution. |
| Assuming heavy users are your best users. | On a flat plan, engagement and cost move together while revenue doesn't. | Check whether your most engaged accounts are profitable before optimising for engagement. |
Run this check tonight
- Multiply each tier's price by your target gross margin to get its cost budget.
- Divide that by your real cost per call to get the calls the tier can afford.
- Count how many customers on that tier are already above it.
- Check whether your highest-usage accounts sit on your lowest-priced plan.
What a $29 tier can afford, and what one heavy user costs
Illustrative example
A modelled $29 monthly tier with a target gross margin of 80%, so a cost budget of $5.80 per customer. Calls are a modelled 3,000 input and 500 output tokens on Claude Sonnet 5 at its real introductory rates of $2.00 and $10.00 per million from the price book effective 2026-07-02. The price, the margin target and the call shape are modelled, the rates are published.
Five hundred and twenty-seven calls is the number that should have been in the pricing discussion, and it takes one division to get. The customer at 5,000 calls isn't abusing anything. They're using the product as advertised, which is the problem with advertising it that way.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| 527 calls | modelled monthly calls a $29 tier can afford at an 80% target gross margin | calculated | A modelled 3,000 input and 500 output tokens per call on Claude Sonnet 5 at real introductory rates of $2.00 and $10.00 per million from the price book effective 2026-07-02 costs $0.0110. A modelled $29.00 tier at an 80% target gross margin leaves a $5.80 cost budget, and $5.80 / $0.0110 = 527 calls. A customer at 5,000 calls costs $55.00, a $26.00 loss against the $29.00 price. Rates are published, the price, margin target, call shape and usage are modelled. |
What a generic answer can’t know
This calculation needs cost and revenue in the same place, and they start in different systems. Model spend arrives on a provider invoice with no customer attached, and what each customer pays lives in your billing system with no tokens attached. Nothing joins them by default, which is why margin per customer is so rarely reported and blended margin so commonly is. Culpa prices every call in exact decimal from a versioned price book, attributes it to the customer and plan that caused it, and takes your revenue alongside, so margin per customer and per tier are queries. That makes the two questions a pricing decision needs answerable from your own accounts: how many customers already exceed their tier's budget, and whether the ones who do are the ones you most want to keep.
Questions founders ask next
How do I work out a tier's usage limit?
Multiply the price by your target gross margin to get the cost budget, then divide by your real cost per call. In the modelled example a $29 tier at 80% target margin has a $5.80 budget, which at $0.0110 a call is 527 calls a month.
Can I offer unlimited usage?
You can, and it means fixed revenue against unbounded cost. That's a deliberate bet rather than a mistake, and the way to take it safely is to know where the line is before anyone crosses it. In the modelled example one customer at 5,000 calls turns a $29 plan into a $26 monthly loss.
Why does blended margin look fine when some customers lose money?
Because usage is skewed. Most customers use a fraction of what they could, which subsidises the tail. Blended margin averages the two and stays comfortable, so the loss-making group is invisible until you look at margin per customer.
Are heavy users bad customers?
On a flat plan they can be, which is uncomfortable because they're usually your most engaged. Engagement raises cost and doesn't raise revenue. That's a pricing problem rather than a customer problem, and it's much cheaper to fix before the plan is public.
On your infrastructure
Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.
How Culpa works
Find the culprit. Not just the total.
Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.
Your prompts stay local.
Culpa runs on your own infrastructure. What you send to a model reaches us at no point.
Every dollar has a name.
Follow any charge to the conversation, the user, the feature and the customer behind it.
See the bill before it lands.
Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.
Three steps to your first answer.
Change one base URL.
Or drop in the Python or TypeScript library.
Find your most expensive conversation.
In the first session, not the first week.
Cost your next feature before you ship it.
Why the bill went up
Example dashboardCalls traced
418,209
across 3 projects
Spend this week
$378.41
+ $182 vs last week
Failed calls
312
74% retried, and you paid for all of them
+ $182 this week traced to one culprit
Spend over 14 days
Most expensive users
Next week forecast
Graded against reality. Accuracy shown as results land.
Keep reading
Sources: Anthropic pricing. Last reviewed 2026-08-05, rates effective 2026-07-02. Plain text version.