OpenAI posts Codex rates
OpenAI published a new Codex rate card that signals vendors are making pricing and entitlements more explicit, which directly affects how platform teams route and bill LLM traffic. The brief said enterprises on legacy plans should keep using older cards for now while migrations to the new rates will happen in the coming weeks, underscoring the need for policy‑aware routing and spend attribution in AI gateways. (help.openai.com)
OpenAI quietly changed how Codex gets billed on April 2, 2026: new plans now pay by tokens, while many older plans still pay by message for now. That means the same coding task can be counted two different ways inside the same vendor’s product line. (help.openai.com) Codex is OpenAI’s coding agent, and it shows up in places like the web app, the command line interface, the software development kit, and the integrated development environment extension. OpenAI says it is included across ChatGPT Plus, Pro, Business, and Enterprise or Education plans, with different limits and billing rules depending on the plan. (help.openai.com) The old system was easy to picture: one request, one charge bucket. The new system works like an electricity meter, where cost depends on how many input tokens, cached input tokens, and output tokens a workspace actually uses. (help.openai.com, developers.openai.com) OpenAI’s help page says the token-based update applies to new and existing ChatGPT Business plans and to new ChatGPT Enterprise plans. The same page says existing Plus, Pro, and Enterprise or Education customers should keep using the legacy rate card until OpenAI migrates them in the coming weeks. (help.openai.com) OpenAI repeated the split in two release-note posts published in late March 2026. The Business notes say Codex now aligns with token-based usage pricing, while the Enterprise and Education notes say new Enterprise workspaces use token-based rates and older Enterprise and Education workspaces stay on message-based rates until migration. (help.openai.com, help.openai.com) That creates a very practical headache for platform teams. If one business unit is on a new workspace and another is on a legacy workspace, an internal gateway has to know which policy to apply before it can estimate cost, enforce limits, or show a chargeback report. (help.openai.com, help.openai.com) OpenAI’s broader pricing pages show why the accounting gets more granular once everything is token-based. The public application programming interface rate sheet now lists separate prices for input, cached input, and output tokens, and the Codex pricing page says application programming interface key usage already follows that token model. (openai.com, developers.openai.com) The company is also separating access from billing more clearly. The Codex pricing page says an application programming interface key gives you Codex in the command line interface, software development kit, or integrated development environment extension, but not cloud features like GitHub code review or Slack, and it warns that access to new Codex models can arrive later on that route. (developers.openai.com) On the team side, OpenAI is pairing the new price logic with new seat logic. Its Enterprise help page says admins can assign seat types and roles, while its flexible-pricing page says Enterprise and Education workspaces buy a shared credit pool at the contract level and can use role-based access control to set spend controls by group. (help.openai.com, help.openai.com) OpenAI’s own launch post frames this as pay-as-you-go pricing for teams that want to start small and scale. Read from the buyer’s side, it is also a sign that model vendors are moving away from fuzzy “included usage” promises and toward line items that finance teams can route, meter, and audit. (openai.com)
