Guides / llm cost spike after launch

Why your LLM cost spikes after a launch

A launch spike is rarely proportional to traffic. New users explore more per session than retained ones, they arrive on your cheapest tier, and any unbounded loop meets its worst input at once. Culpa, a local-first LLM cost, margin, and forecast ledger, forecasts a launch from your own usage curve and prices it before you ship.

Why this happens

Three effects stack on launch day. A curious first-time user sends more messages per session than a habitual one, so cost per user runs above your steady-state figure. Launch traffic skews heavily toward free and trial accounts, so the expensive cohort is also the one paying least. And volume finds the edge cases your loops never hit at low traffic, which is when an unbounded retry or agent run becomes a real number.

What this usually looks like

Spend rises faster than signups, so cost per new user exceeds your modelled figure.
The spike doesn't decay for several days, because exploration continues past day one.
Your worst single conversation of the month happens during launch week.
Free-tier accounts account for most of the increase while contributing no revenue.

Free, no card, no account

Run the free Cost Leak Scan

It shows your most expensive conversation before you install anything.

Run the free Cost Leak Scan Start 14-day trial

Mistakes that cost the most

Mistake	Why it hurts	Do instead
Forecasting a launch by multiplying steady-state cost per user by expected signups.	New users cost more than retained ones, so the forecast is low exactly when being low matters.	Use first-week cost per user from a previous cohort, not the blended figure across all users.
Shipping without a per-account spend cap.	Launch traffic is where a single account can generate an unbounded bill before anyone notices.	Set a per-account daily ceiling before launch, and alert on approach rather than on breach.
Publishing a single forecast number for launch month.	Launch variance is the widest of any month, so a point estimate is read as a commitment and then missed.	Publish a range and widen it deliberately for launch week, stating the assumption behind each bound.

Run this check tonight

Pull cost per user for their first seven days from a past cohort, and compare it to your blended figure.
Check whether a per-account spend ceiling exists in code today. If not, your launch worst case is unbounded.
Identify which model your onboarding flow uses. It's often the most expensive one, on the least valuable traffic.
Price your single most expensive conversation ever, then multiply by a plausible number of curious new users.
Confirm your alerting fires on spend rate, not only on daily total. A daily total tells you the morning after.

Launch week priced from a first-week cohort, not the blended average

Illustrative example

2,000 signups in launch week. Steady-state users cost $0.22 a month on GPT-5.4 at real rates, $0.0025 per 1k input and $0.015 per 1k output. First-week users are modelled at 3.2 times the steady-state token usage.

Naive forecast: 2,000 x $0.22 = $440 for the month

First-week usage at 3.2x steady state = $0.22 x 3.2 = $0.70 per new user in week one

Launch week alone: 2,000 x $0.70 = $1,400

Remaining three weeks at steady state: 2,000 x $0.22 x (3/4) = $330

Realistic month = $1,400 + $330 = $1,730, against a naive $440

The naive forecast under-reads by nearly four times, entirely because it priced new users as if they behaved like retained ones. The fix is a cohort number, not a bigger multiplier.

Every number, with its confidence and source

Figure	What it means	Confidence	Source
$440 to $1,730	modelled launch-month spend, naive forecast versus cohort-based forecast	estimated	Both endpoints from the teardown arithmetic at real GPT-5.4 rates from the price book, effective 2026-07-02. A range because the first-week usage multiple is modelled, not measured.
$0.22	modelled steady-state model cost per active user per month	calculated	(40 x $0.0025) + (8 x $0.015) using GPT-5.4 rates per 1k tokens from the price book. Per-user token volumes are modelled.

What a generic answer can’t know

A provider has no idea you launched. It can't separate a first-week user from a habitual one, because it has never seen your signup dates or your plan tiers. That join lives in your own data. Culpa reads it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.

Questions founders ask next

How much more do new users cost than retained ones?

It varies by product and you should measure your own rather than borrow a figure. The direction is consistent though: first-week usage runs above steady state because exploration is front-loaded, so a blended average always under-forecasts a launch.

Should I cap usage during a launch?

Cap per account rather than globally. A global cap degrades the launch for everyone when one account misbehaves, and the whole point of a launch is that the experience holds up.

How long does a launch spike last?

Longer than most teams budget for, because exploration continues past day one and the cohort decays gradually rather than dropping. Model it as a decay over the first weeks, not a single-day peak.

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.

Base URLhttp://localhost:4545/v1Your traffic keeps flowing if Culpa ever stops.

Why the bill went up

Example dashboard

Calls 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

user_384report_generatorconv_91fprompt_v1894,220 tokens3 retries$6.81

Most expensive users

user_384$38.42

user_119$21.07

user_562$14.90

user_204$8.30

user_871$5.10

Next week forecast

Best$180

Median$240

p90$310

p99$395

Graded against reality. Accuracy shown as results land.

Run the free Cost Leak Scan Start 14-day trial

Keep reading

Sources: OpenAI API pricing. Last reviewed 2026-08-01. Plain text version.