OpenAI formalizes Codex pricing

OpenAI has stopped treating Codex like a side feature and started pricing it like metered infrastructure. In an update dated April 2, 2026, the company published a formal Codex rate card and said new ChatGPT Business and new Enterprise customers now pay based on token usage instead of per-message estimates. (help.openai.com) That change means Codex is being billed more like cloud computing than like a chatbot subscription. OpenAI’s new table lists separate credit charges per 1 million input tokens, cached input tokens, and output tokens, with GPT-5.4 output priced at 375 credits per 1 million output tokens and GPT-5.1-Codex-mini output priced at 50 credits. (help.openai.com) A token is a small chunk of text, so this system charges for how much code and context the agent actually processes. OpenAI says the token-based format replaces average per-message estimates with a direct mapping between usage and credits, and fast mode burns credits at 2 times the normal rate. (help.openai.com) OpenAI also folded Codex deeper into the main ChatGPT lineup instead of keeping it as a separate developer lane. The current pricing page says Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Education, and Enterprise plans, with Plus at $20 a month and Pro at $200 a month. (developers.openai.com) The split between plans shows who OpenAI thinks Codex is for. Plus is pitched for “a few focused coding sessions each week,” while Pro is pitched for “daily full-time development,” and Business adds admin controls, larger virtual machines, and usage-based seats for teams. (developers.openai.com) OpenAI is doing this while demand is spiking fast. Multiple reports on April 8 and April 9 said Codex had reached about 3 million weekly users, and Sam Altman said OpenAI reset usage limits after the milestone, with plans to repeat that at each additional million users until 10 million. (technobezz.com) (msn.com) That is the same pattern cloud companies follow when a product stops being experimental and starts straining capacity. You publish a rate card, meter usage more precisely, and loosen or tighten limits depending on how many people are hitting the system at once. (help.openai.com) (technobezz.com) The next step is money flowing through the apps these agents build. On April 6, Razorpay said developers using Codex can connect Razorpay inside the agent so an app can be generated and payment collection can be set up almost instantly through Razorpay’s Model Context Protocol server. (business-standard.com) Razorpay’s example was blunt: a developer can ask Codex to build an artificial-intelligence fitness coaching app and wire up checkout flows at the same time. That turns Codex from a code-writing assistant into something closer to a one-person software shop that can build the product and prepare it to charge customers in the same workflow. (business-standard.com) OpenAI’s own help page now says average Codex usage runs about $100 to $200 per developer per month, with big variation based on model choice, automations, number of instances, and fast mode. That is not hobby pricing anymore; it is the kind of spend companies budget for tools they expect employees to use every day. (help.openai.com) The picture that emerges is simple: Codex is being packaged like a core product, billed like compute, and connected to payments so the software it generates can start earning money faster. Once a tool has a formal rate card, plan tiers, admin controls, and checkout built into the workflow, it is no longer a demo. (help.openai.com) (developers.openai.com) (business-standard.com)