Use case
GPT-5.4 Pro prompt caching
prompt caching notes for gpt-5.4-pro: the rate card, a worked volume, and when to send this ID. A sample of 1 million input tokens plus 200 thousand output tokens is about $13.20 on corouter (input $6.00, output $7.20).
Cache-read rate
Cache read for gpt-5.4-pro lists at $3.00 and is $0.600 on corouter per million tokens. Uncached input is $6.00.
Only the input that actually hits cache is billed at the cache-read rate. This page does not promise your prompt will hit.
Where this ID sits
GPT-5.4 Pro is the outlier on the rate card: its list output rate is far above GPT-5.4. Send gpt-5.4-pro only when you have already chosen that ID on purpose.
Questions
- Which model ID should the request send?
- Send gpt-5.4-pro. Do not put a different name in the model field.
- What is the minimum top-up?
- Prepaid credits, $5 minimum top-up, no subscription tier.