Codex token pricing explained (and how to pay less)
Token pricing looks simple — a price per million input and output tokens — but a few details decide your real bill.
Input vs. output
Output tokens typically cost several times more than input. Verbose answers and long code dumps are where money goes.
Context re-sending
Every turn in a conversation re-sends prior context. Long sessions compound.
Cached input
Where the provider supports prompt caching, repeated prefixes are billed at a reduced rate. Structure prompts so stable instructions come first.
The reseller route
Discount resellers like Token Supply offer the same per-token model at a lower rate — up to 70% off on some models. For high-volume, price-sensitive workloads this usually dwarfs every other optimization.
Check exact model availability and rate limits before moving production traffic.