OpenAI posts Codex pricing and changelog

OpenAI just did something software companies usually avoid doing: it posted a rate card that says exactly how Codex usage is billed, down to credits per 1 million input tokens, cached tokens, and output tokens. On April 2, 2026, OpenAI said new and existing ChatGPT Business customers and new ChatGPT Enterprise customers move to token-based Codex pricing instead of per-message pricing. (help.openai.com) That changes the unit you manage. A “message” is like paying one flat fare for a taxi ride, while a “token” price is like paying by the mile, with separate lines for what you send in, what gets reused from cache, and what the model writes back. (help.openai.com) OpenAI’s table makes that concrete. For GPT-5.4 in Codex, the posted rate is 62.50 credits per million input tokens, 6.250 credits for cached input tokens, and 375 credits for output tokens, while GPT-5.1-Codex-mini is listed at 6.25, 0.625, and 50 credits respectively. (help.openai.com) The cache line is the part finance teams will stare at. Cached input tokens are reused context, so OpenAI prices them at about one-tenth of normal input cost in the examples on the page, which rewards teams that keep repeating the same codebase context instead of resending everything from scratch. (help.openai.com) There is also a speed surcharge. OpenAI says Fast mode consumes 2 times as many credits, and the same page says average Codex cost runs about $100 to $200 per developer per month, with large variation based on model choice, parallel instances, automations, and Fast mode usage. (help.openai.com) The rollout is not one clean switch for every customer. OpenAI says Plus and Pro users, along with existing Enterprise and Edu customers, stay on a legacy message-based Codex rate card until migration in the coming weeks, while new Enterprise workspaces and Business customers use the new token-based rates now. (help.openai.com 1) (help.openai.com 2) OpenAI also split access into seat types for business customers. On April 2, 2026, it introduced a Codex seat for ChatGPT Business and ChatGPT Enterprise that has no fixed monthly seat price, gives access to Codex only, and requires workspace credits because billing is based on usage. (help.openai.com 1) (help.openai.com 2) That matters because a company can now buy one kind of seat for general chat and a different kind for coding work. OpenAI says standard ChatGPT seats in Enterprise still include ChatGPT and Codex with baseline access, while Codex-only seats are a separate flexible-pricing option for members who do not need the full workspace product. (help.openai.com) The other half of the story is the paper trail. OpenAI’s ChatGPT release notes page, updated April 10, 2026, keeps logging dated product changes such as ChatGPT in Apple CarPlay on April 2, updated Box, Notion, Linear, and Dropbox apps on March 27, and a simplified mobile sidebar on March 26 that puts experiences like Images, Codex, Pulse, and Apps above chats and projects. (help.openai.com) That changelog is not just marketing copy if you run budgets or procurement. When a vendor publishes dated release notes, separate business-plan release notes, enterprise release notes, and a model-by-model rate card, buyers can compare one supplier’s coding agent the way they compare cloud storage or database usage: by unit cost, feature history, and who gets what on which date. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3)