Guides / gpt-5.5 pricing
GPT-5.5 pricing, and when the premium earns itself
GPT-5.5 pricing is $5.00 per million input tokens and $30.00 per million output tokens, exactly double GPT-5.4 on both sides. The premium earns itself only when a wrong answer costs more than the call. Culpa, a local-first LLM cost, margin, and forecast ledger, meters both models side by side so the comparison is measured, not argued.
Why this happens
GPT-5.6 Sol now bills at this same $5.00 and $30.00, so staying on GPT-5.5 saves nothing at all and the choice between them is a behaviour question rather than a cost one. Against GPT-5.4 the rate is a clean 2x, which makes that decision unusually easy to reason about. The question isn't whether it's better, it's whether the improvement is worth doubling that step's cost. Teams almost never answer that with data. They route everything to the stronger model after one bad output from a cheaper one, and the mix never gets rolled back because nobody owns the review.
What this usually looks like
- Traffic migrated to GPT-5.5 after a single quality incident and never migrated back.
- Your spend doubled on a feature whose behaviour nobody can demonstrably show improved.
- The share of calls on GPT-5.5 keeps rising with no decision recorded anywhere.
- Retries escalate to GPT-5.5 automatically, so your most expensive model handles your most failure-prone work.
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 |
|---|---|---|
| Switching to GPT-5.5 after one bad output. | You double the cost of every call to fix a failure rate you never measured. | Measure the failure rate on both models over a week, then price the difference against the doubling. |
| Escalating retries to GPT-5.5 by default. | Failures cluster, so the escalation path carries far more traffic than the happy path suggests. | Escalate only for error classes a stronger model actually fixes, and track how often it fires. |
| Running the whole pipeline on GPT-5.5 because one step needs it. | You pay double on the formatting and routing steps that would run identically on a cheap model. | Split the pipeline. Put GPT-5.5 on the one step that needs it and measure the rest separately. |
Run this check tonight
- Compute what share of your tokens run on GPT-5.5 today, and what it was a quarter ago.
- Price the same volume at GPT-5.4 rates. The difference is what the upgrade costs you monthly.
- Measure the failure rate of both models on a hundred real inputs from your own traffic.
- Check whether any retry path escalates to GPT-5.5, and count how often it fires.
- Ask what a wrong answer actually costs on this surface. If the answer is nothing much, the premium is hard to justify.
The break-even on doubling your rate
Illustrative example
A feature sending 5 million input and 1 million output tokens a month. The question is whether moving from GPT-5.4 to GPT-5.5 pays for itself, given a modelled failure rate that a human then has to correct.
$27.50 a month buys the upgrade on this volume. Whether that's cheap depends entirely on what a wrong answer costs you, which is a number only you have.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $5.00 / $30.00 per million | GPT-5.5 input and output rates | calculated | Culpa price book row for openai/gpt-5.5, $0.005 and $0.030 per 1k, effective 2026-07-02, verified against OpenAI API pricing. Shown here per million. |
| 2.0x | GPT-5.5 cost relative to GPT-5.4 on both input and output | calculated | $5.00 / $2.50 and $30.00 / $15.00, from the two price-book rows. |
What a generic answer can’t know
A rate card can show you that GPT-5.5 is double GPT-5.4. It can't show you your failure rate on either, because failure is a judgement your application makes after the response arrives. Comparing them needs both sides of the call. Culpa reads that on your infrastructure, where your prompts and responses stay, and counts the calls to run your plan.
Questions founders ask next
How much more does GPT-5.5 cost than GPT-5.4?
Exactly double on both sides. $5.00 against $2.50 for input, $30.00 against $15.00 for output, per million tokens. That clean ratio makes the trade unusually easy to model against your own failure rate.
When is GPT-5.5 worth the premium?
When a wrong answer costs more than the doubled call. Work that a human reviews and corrects has a real per-failure cost, so the arithmetic closes quickly. Work nobody checks rarely justifies it.
Does GPT-5.5 have a cached rate?
Yes, at $0.50 per million, a 90% discount against its $5.00 input rate. The same rule applies as anywhere else, the repeated block has to sit at the front of the prompt and stay identical or the discount never lands.
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.
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.