Guides / gpt-5.4 nano pricing
GPT-5.4 nano pricing, and the ladder it sits at the bottom of
GPT-5.4 nano pricing is $0.20 per million input tokens and $1.25 per million output tokens, which is 150 times below the Pro tier's $30.00 input on the same rate card. The widest cost decision you'll make sits inside one vendor rather than between two. Culpa, a local-first LLM cost, margin, and forecast ledger, prices each call by the model that served it.
Why this happens
Model selection gets framed as a choice between providers, and the arithmetic says otherwise. OpenAI's own ladder runs from $0.20 to $30.00 per million input, a spread of 150 times, which is far wider than the gap between any two providers at the same tier. So the question that moves your bill isn't which vendor you signed with. It's which rung each request lands on, and most stacks route everything to a middle rung because that decision was made once and never revisited.
What this usually looks like
- Every request goes to the same model regardless of how hard the task is.
- Nobody has tested whether nano handles your classification and routing work.
- The tier was chosen during a quality push and has never been re-examined.
- Cost per request is uniform across tasks that differ enormously in difficulty.
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 |
|---|---|---|
| Comparing vendors before comparing tiers within one vendor. | The within-vendor spread here is 150x, and no cross-vendor comparison at the same tier comes close. | Price your workload on every rung of one ladder first. The answer is usually there. |
| Routing mechanical work to a middle tier by default. | Classification and extraction rarely need it, and nano runs at a twelfth of the standard tier's input rate. | Split traffic by task type and send the mechanical share down, with a check on the output. |
| Ignoring the cached rate because the model is already cheap. | Cached input here is $0.02 per million, a 90% discount, and percentages still apply at small rates. | Order the prompt stable-first even on the cheap tier, then confirm cached tokens are billing. |
Run this check tonight
- Price last month's workload on nano, mini, standard and Pro. Read the four numbers together.
- Sort your request types by difficulty and mark which genuinely need more than nano.
- Sample 200 real requests, run them through nano, and score the failures rather than assuming them.
- Check what share of nano's input bills at the cached $0.02 rate.
- Compare nano against GPT-5.6 Luna, which matches its input rate and prices output slightly lower.
One workload, four rungs of the same ladder
Illustrative example
A workload of 50 million input and 10 million output tokens in a month, priced on four OpenAI tiers at real rates. Volumes are modelled and the ratios hold at any volume.
The gap between the top and bottom of one vendor's ladder is 147 times on identical work. Any comparison that starts with which provider to use has skipped the decision that actually sets the bill.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $0.20 / $1.25 per million | GPT-5.4 nano input and output rates | calculated | Culpa price book row for openai/gpt-5.4-nano, $0.0002 and $0.00125 per 1k, effective 2026-07-02, verified against OpenAI API pricing. Shown here per million. |
| 150x | input rate of the Pro tier as a multiple of nano, on the same OpenAI rate card | calculated | $30.00 divided by $0.20, from the gpt-5.4-pro and gpt-5.4-nano price-book rows. |
| $22.50 to $3,300.00 | modelled monthly cost of one identical workload across four OpenAI tiers | estimated | Both endpoints from the teardown arithmetic at real OpenAI rates. A range because the token volumes are modelled. |
What a generic answer can’t know
A rate card lists four tiers. Which of your requests each tier can actually handle is a fact about your own prompts and your own quality bar, and it only comes out of running them. Culpa prices each call by the model that served it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.
Questions founders ask next
How much does GPT-5.4 nano cost?
$0.20 per million input tokens and $1.25 per million output, on rates verified against OpenAI's pricing page. Cached input runs at $0.02 per million, a 90% discount on the standard input rate.
What is nano actually good for?
High-volume tagging, routing, classification and guardrail checks, where latency and unit cost matter more than nuance. The way to find out is to run a sample of your real requests through it and score the failures rather than predicting them.
How does nano compare to GPT-5.6 Luna?
They match at $0.20 per million input, and Luna prices output slightly lower at $1.20 against $1.25. On an output-heavy workload that small gap compounds, so it's worth checking which one your traffic favours.
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.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Keep reading
Sources: OpenAI API pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.