The Codex Credits Rate Card, Explained

Credits are how Codex bills usage beyond a plan's included allowance, and how credit-based Business, Enterprise and Edu workspaces pay for it. OpenAI publishes a rate per million tokens for each model — uncached input, cached input and output — and a formula that multiplies each token type by its rate and adds them up. The tables above list every row, the formula, and the speed multipliers.

Credits track API prices

The third table divides each model's credit rate by its own API price, and the result comes out the same for every model that has both. In other words, the rate card is the API price list expressed in a different unit — which makes it easy to compare models in credits using everything you know about them in dollars. It doesn't tell you what a credit costs you: OpenAI says credit prices and discounts depend on your plan or agreement, so this site never assumes one.

No cache-write charge

The formula has no cache-write term. OpenAI states that Codex credit billing has no separate cache-write charge, unlike API usage on GPT-5.6 and later models, where written tokens bill at their own higher rate. For workloads that write a lot of new cache — short sessions, frequently changing prefixes — that's a real difference between paying in credits and paying per token.

Speed multiplies credits, and multiplies included usage more

Fast multiplies credits by 2x; GPT-6 Astra Ultrafast by 6x. Against a plan's included usage the same speeds count at 2.5x and 8x. OpenAI notes these are billing multipliers, not speed figures.

A sense of scale

OpenAI's own estimate is that a typical GPT-5.6 Sol task uses 5–30 credits — a wide range, because task size dominates. For your own tasks, read the token counts from /status and use the Codex credits calculator, which also finds the credit price at which paying with credits beats an API key.

Rows without an API model

A few rows — Daybreak Blue and Red, GPT-Rosalind-Research, and the image rows — have no matching API model on this site, so they appear here and in the calculator but not in API-price comparisons.