Codex Credits vs API Billing: ChatGPT Work and API Keys

What a credit buys on each GPT-6 model

The table sets OpenAI's credit rates at Standard speed beside the API's standard short-context dollar rates (prompts up to 272K input tokens) for the same model. Read across any row: each credit rate is exactly 25 times the API dollar rate, for input, cached input and output alike. So one credit buys the tokens that $0.04 buys on the API.

Illustrative only: a turn that sends 40,000 input tokens and returns 2,000 output tokens, the size used in Codex: ChatGPT login vs API key, uses 2.5 credits on GPT-6.1 Sol (2 for input, 0.5 for output) and 0.125 credits on GPT-6 Luna. The same turn costs $0.10 and $0.005 on an API key. The older GPT-5.6 models have their own credit rates; OpenAI says a typical GPT-5.6 Sol task may use 2 to 15 credits.

Two details differ from the API. Codex credit billing has no separate cache-write charge, while the API bills cache writes at 1.25 times the input rate. Faster speeds raise both bills by the same factor: Fast doubles the purchased-credit rate, as the API's Fast tier doubles the dollar rate, and Ultrafast on GPT-6 Astra and GPT-6.1 Sol is six times on both. Included plan usage runs down faster than that, at 2.5 times on Fast and 8 times on Ultrafast.

No published credit price, so no dollar break-even

OpenAI says credit purchase prices and discounts depend on your plan or agreement, and its Codex pricing page lists none. The published numbers therefore cannot give a dollar break-even between credits and an API key. They do give a test you can run with your own price: if a credit costs you less than 4 cents, credits buy the same tokens more cheaply than an API key; above 4 cents, the API key is cheaper. Long sessions that keep writing new context to the cache tilt slightly toward credits, which skip the cache-write charge. Above 272K input tokens the API doubles input and raises output by half; OpenAI's credit table does not say whether credits do the same.

The rates also do not tell you how much a plan includes. OpenAI says credit prices alone don't determine included subscription usage, and it warns against using API prices to estimate included tasks.

Workspace credits on Business, Edu and Enterprise

Business, Edu and Enterprise plans with flexible pricing can buy workspace credits for the whole workspace. On Enterprise and Edu with flexible pricing there are no fixed rate limits, and usage scales with the credits the workspace holds. Enterprise and Edu without flexible pricing have the same per-seat limits as Plus for most features. Which arrangement your workspace has decides whether more Codex usage is a credit purchase at all, so ask your admin before budgeting for it.

At the limit: buy credits or sign Codex in with an API key

When a plan runs out, any user can also keep running local chats with an API key at standard API rates. The table shows what changes with each choice.

The API key wins on a known dollar price and on the training default for personal plans. Credits are the only top-up for work that needs Codex cloud, GitHub code review, Slack or voice, because an API key cannot reach them.

When to stop topping up and price Pro

On Plus, the point to stop topping up is about $80 a month. That is the gap between Plus at $20 and the lowest Pro plan at $100, which currently has no five-hour limit. At the illustrative turn size, $80 buys 800 GPT-6.1 Sol turns on an API key, while 1,000 GPT-6 Luna turns cost $5, or 125 credits. For credits, add up what you paid in a month; once it nears $80, price Pro instead. ChatGPT Plus vs Pro covers what else that step buys.

For the terminal costs of each sign-in, see Codex CLI cost. For per-plan limits, see ChatGPT Work and Codex limits by plan.