Codex Credits Calculator

Signed in with ChatGPT, Codex doesn't bill you per token in dollars. Work beyond your plan's included usage is paid for in credits, at a published rate per million tokens that differs by model, with speed modes multiplying it. This calculator turns the token counts of a real Codex task into credits, shows what the same task costs on every model with a published credit rate, and — because a credit's money value depends on your plan — tells you the credit price at which paying with credits beats running the same tokens on an API key.

What you enter

The model, as a row from OpenAI's credit rate card; the speed; and the task itself as uncached input tokens, cached input tokens and output tokens, with reasoning counted inside output. Tasks per day turns one task into a daily and monthly total. The last field is optional: what one credit costs you. It starts empty, because OpenAI says credit prices and discounts depend on your plan or agreement, and a default would be wrong for most readers. Leave it blank and the tool still works; fill it in and the monthly total gains a dollar figure.

Reading real token counts is the part people skip. Run a representative task, then check /status in the CLI for its token usage, rather than guessing — a task that re-sends a large context every turn is mostly cached input, and that changes the answer more than the choice of model does. Checking Codex usage with /status and /usage walks through it.

How the number is built

Each token type is multiplied by its row's rate per million and summed — the formula OpenAI publishes. There is no cache-write term, because Codex credit billing has no separate cache-write charge, unlike the API on GPT-5.6 and later models. Fast multiplies the credits charged by 2x; Ultrafast, offered only for GPT-6 Astra, by 6x. The speed menu only offers what the selected model supports, so you can't price an Ultrafast task on a model that has none.

Included usage burns faster than credits

When you pick a faster speed, the result also shows the task in included-usage terms, and the multipliers there are steeper: Fast counts against your plan's allowance at 2.5x the Standard rate, and Astra Ultrafast at 8x. OpenAI notes that these are billing multipliers, not speed figures. The practical consequence is that the same Fast session drains a plan's allowance proportionally faster than it drains a credit balance — why Fast mode burns your Codex plan faster works through what that means for when to switch it on.

Comparing every model on the same task

Below the result, the same token counts are priced on every row of the rate card at the selected speed. That's the quickest way to see what a model change does to a credit budget: credit rates track API prices closely, so the spread between the cheapest and most expensive rows is wide, and a task that is routine on one model can be a significant spend on another. Rows that don't offer the selected speed say so rather than showing a number.

The break-even against an API key

For any model with a published API price at the selected speed, the tool prices the same tokens on an API key and divides by the credits charged. The result is the credit price at which both ways of paying cost the same: if a credit costs you less than that, credits are cheaper for this task; if more, the API key is. Enter your own credit price and the tool says which side of the line you're on.

A couple of caveats keep that comparison honest. It assumes no single request in the task crosses the long-context threshold, because API pricing reprices a whole request past 272K input tokens and credit billing doesn't publish an equivalent rule. And it compares billing only: a ChatGPT plan's included usage, its features and its limits are part of what you pay for, and none of that is a per-token figure.

What the tool doesn't assume

It never fills in a credit price, a plan price or a typical task size for you. The default token counts exist so the page shows a working example; replace them with a task of your own. Where OpenAI publishes an estimate — local messages per 5-hour window on Plus for each model — the tool shows it for scale, labeled as OpenAI's estimate rather than a limit, because OpenAI says it isn't one.

Where to go next

The Codex credits rate card lists every rate this tool uses, along with the formula and the speed multipliers. Codex plans and usage limits covers what each plan includes before credits come into play, and the subscription vs API cost calculator frames the bigger question of whether a plan is the right way to pay at all.