Guides / gpt-5.4 mini pricing

GPT-5.4 mini pricing, and where the saving disappears

GPT-5.4 mini pricing is $0.75 per million input tokens and $4.50 per million output tokens, roughly a third of GPT-5.4 on both sides. Cheap enough that teams stop counting calls, which is how volume eats the saving. Culpa, a local-first LLM cost, margin, and forecast ledger, meters every call so a cheap model can't hide behind its rate.

Why this happens

Mini exists to take the mechanical work off the expensive model, and it does that well. The failure mode is behavioural rather than technical. A rate three times lower feels like a rounding error, so nobody instruments the feature, retry caps get relaxed, and prompt length stops being reviewed. Then call volume runs into the millions and the cheapest model on the roster becomes a line item nobody can explain.

What this usually looks like

  • A mini-powered feature appears in your top three by spend, which nobody expected.
  • Mini traffic carries no feature tag, because it seemed too cheap to be worth instrumenting.
  • Retry caps on mini paths are looser than on frontier paths, so failures multiply against a huge call count.
  • Prompt length on mini calls has never been reviewed, because the per-call cost looks negligible.

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 ScanStart 14-day trial

Mistakes that cost the most

MistakeWhy it hurtsDo instead
Skipping attribution on mini calls because the rate is low.Volume is the multiplier here, and untagged spend is exactly where an unexplained line item grows.Tag mini traffic the same way you tag frontier traffic. The cost of the tag is nothing.
Copying a frontier retry policy onto mini paths.Mini handles very high call counts, so a five-attempt policy multiplies an enormous denominator.Cap retries at two and measure attempts per successful result, the same as anywhere else.
Leaving long system prompts on mini calls.At millions of calls a month, a thousand wasted prompt tokens is a real monthly number.Price the prompt: call volume times prompt tokens times $0.75 per million. Then decide.

Run this check tonight

  1. Find your monthly mini call count. It's usually an order of magnitude above your frontier count.
  2. Multiply prompt tokens by that call count by $0.75 per million. That's your fixed prompt cost.
  3. Check whether mini traffic carries feature tags. If not, it's sitting in unattributed spend.
  4. Compare attempts to successful results on mini paths specifically.
  5. Confirm mini is actually doing mechanical work, not reasoning work it fails at and then retries.

Where a third of the rate stops being a third of the bill

Illustrative example

A routing step on mini handling 8 million calls a month. Each call sends 900 input tokens, mostly a fixed system prompt, and returns 60 output tokens.

Input: 8,000,000 x 900 = 7,200M tokens x $0.75 per million = $5,400
Output: 8,000,000 x 60 = 480M tokens x $4.50 per million = $2,160
Month total = $7,560 for a step everyone describes as cheap
Trimming the prompt from 900 to 400 tokens: 3,200M x $0.75 = $2,400, saving $3,000
The same work on GPT-5.4 would be (7,200 x $2.50) + (480 x $15.00) = $18,000 + $7,200 = $25,200

Mini saves $17,640 a month against GPT-5.4 here, and still costs $7,560. Trimming five hundred prompt tokens saves another $3,000, which is more than most teams find anywhere else.

Every number, with its confidence and source

FigureWhat it meansConfidenceSource
$0.75 / $4.50 per millionGPT-5.4 mini input and output ratescalculatedCulpa price book row for openai/gpt-5.4-mini, $0.00075 and $0.0045 per 1k, effective 2026-07-02, verified against OpenAI API pricing. Shown here per million.
$4,560 to $7,560modelled monthly cost of a mini routing step at eight million calls, trimmed prompt versus untrimmedestimatedFrom the teardown arithmetic at real GPT-5.4 mini rates. Published as a modelled scenario because the call volume and token sizes are assumptions.

What a generic answer can’t know

A rate card can't tell you that your cheapest model carries your largest call count, or that most of its input is a system prompt nobody has read in six months. That needs per-call token data joined to your features. Culpa records it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.

Questions founders ask next

How much cheaper is GPT-5.4 mini than GPT-5.4?

About a third on both sides. $0.75 against $2.50 for input and $4.50 against $15.00 for output, per million tokens. The saving is real, but it applies per token, so it disappears if the move to mini comes with looser prompt discipline.

What work should stay on the larger model?

Anything where a wrong answer is expensive or hard to detect. Mini is well suited to routing, classification, extraction and formatting. Test on your own data rather than assuming, because the boundary sits in a different place for every product.

Does GPT-5.4 mini support cached input?

Yes, at $0.075 per million against the $0.75 input rate, the same 90% discount the larger models carry. At mini's typical call volume that discount is worth more in absolute terms than it's on a frontier model.

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.

1

Change one base URL.

Or drop in the Python or TypeScript library.

2

Find your most expensive conversation.

In the first session, not the first week.

3

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

$0$20$40$60$8024262830020406
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.

Keep reading


Sources: OpenAI API pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.