OpenAI formalizes Codex credits pricing

OpenAI just made Codex pricing legible in the way finance teams like. Not simpler, exactly — but formal. The company now has a published Codex rate card that converts usage into token-based credits across consumer and enterprise plans, and it pairs that with a broader push toward pay-as-you-go seats for teams. ### What actually changed? The big change is that Codex is no longer framed mainly as “you get some usage with your plan.” OpenAI says it switched Codex pricing on April 2, 2026 for Plus, Pro, Business, and new Enterprise customers, then extended that on April 23 to existing Enterprise plans, including Edu, Health, Gov, and ChatGPT for Teachers. The new system maps usage to credits per million input, cached input, and output tokens. ### Why does token pricing matter? (help.openai.com) Because per-message pricing hides what’s really expensive. A short prompt with a huge codebase in context is not the same as a tiny local edit, and OpenAI is now saying that directly. The rate card gives separate prices for input, cached input, and output, which means teams can finally see what long context windows, repeated context, and large generated patches do to spend. ### What are the actual numbers? (help.openai.com) For GPT-5.5, Codex is priced at 125 credits per 1 million input tokens, 12.5 credits for cached input, and 750 credits for output. GPT-5.4 is 62.5, 6.25, and 375. GPT-5.3-Codex and GPT-5.2 sit at 43.75, 4.375, and 350. OpenAI also notes that Fast mode costs more, and that GPT-5.3-Codex-Spark is still a research preview with non-final rates. ### Is this just for enterprises? No — that’s the interesting part. The rate card covers Plus and Pro too, and OpenAI’s separate credits doc says those users can buy extra credits once they hit included limits. (help.openai.com) Auto top-up is available for eligible Plus and Pro users, and the same purchased credits can also be used across supported features like ChatGPT for Excel. Basically, OpenAI is normalizing overage spending even for individual subscribers. ### What changed for teams? OpenAI paired the rate card with a product change on April 2: ChatGPT Business and Enterprise workspaces can add Codex-only seats with pay-as-you-go pricing and no fixed seat fee. That gives companies a way to run a narrow pilot — say, a few engineers using Codex heavily — without buying full ChatGPT seats for everyone. OpenAI also cut the annual ChatGPT Business seat price from $25 to $20. ### Why is that a bigger deal than it sounds? (help.openai.com) Because this is procurement language, not toy language. A metered seat, a rate card, usage monitoring, auto-reload credits, and budget visibility all push Codex into the same buying motion as cloud infrastructure. OpenAI even says teams can monitor remaining credit, purchase more, and manage auto-reload from the usage panel. That is the stuff finance and IT admins need before a pilot becomes standard tooling. (openai.com) ### Is adoption actually there? OpenAI says yes. It says more than 2 million builders now use Codex every week, and that Codex usage inside Business and Enterprise has grown 6x since January 2026. It also pegs average Codex cost at roughly $100 to $200 per developer per month, while warning that the range varies a lot by model, automation, and fast-mode use. ### So what’s the bottom line? OpenAI didn’t just publish prices. It turned Codex into something companies can meter, cap, and budget. (help.openai.com) That sounds boring — but it’s usually the moment a product stops being an experiment and starts becoming infrastructure. (openai.com)