OpenAI moves to usage billing

OpenAI’s chat product is starting to look less like Microsoft Office and more like Amazon Web Services. A help page updated this week shows Codex priced in credits across ChatGPT Plus, Pro, Business, and Enterprise or Edu, which is the language of metered compute, not flat software seats. (help.openai.com) That shift matters because OpenAI still sells the main ChatGPT plans per user per month. The public pricing page lists Plus, Pro, Business, and Enterprise as seat-based subscriptions, even as newer features begin to run on separate usage meters inside those plans. (openai.com) Codex is OpenAI’s coding agent, which means it does work that can keep running after you click once. OpenAI’s Codex help collection describes it as a tool that helps users write, review, and ship code faster, which is exactly the kind of product that is hard to bundle into one unlimited monthly fee. (help.openai.com) OpenAI is now saying that Business workspaces can mix two seat types in the same company account. A Business frequently asked questions page says that, starting April 2, 2026, a workspace can include standard ChatGPT seats with a fixed monthly cost, usage-based Codex seats, or both together. (help.openai.com) The company is making the same move higher up the stack for enterprise accounts. OpenAI’s Enterprise and Edu release notes say new ChatGPT Enterprise workspaces use token-based rates, while existing Enterprise and both new and existing Edu, Teacher, and Healthcare workspaces stay on older message-based rates until migration. (help.openai.com) OpenAI has also published a separate rate card for ChatGPT models and features inside Business and Enterprise or Edu. That help page says flexible pricing covers model use and advanced features, while the Codex rate card sits beside it as its own meter, which makes the whole product line read more like cloud infrastructure with multiple billable resources. (help.openai.com) The feature rollout pattern points in the same direction as the pricing pages. OpenAI’s general ChatGPT release notes say Enterprise and Edu users can now choose from the full set of models when building custom Generative Pre-trained Transformers, and that this capability had launched to Plus, Pro, and Team users earlier in the month. (help.openai.com) That means OpenAI is no longer separating tiers only by how many messages a person can send. It is also separating them by what the product is allowed to do, with custom Generative Pre-trained Transformers, apps, actions, admin controls, and model choice arriving on different schedules for prosumers, teams, and regulated organizations. (help.openai.com 1) (help.openai.com 2) You can see the logic in OpenAI’s own description of flexible pricing. The company says credits unlock extra access to Deep Research, reasoning models, image generation, Advanced Voice, and Codex, and Business users can fall back to a shared credit pool after they hit per-seat limits if the workspace owner has bought credits. (help.openai.com) This is the same playbook cloud companies use when a product has cheap usage at the low end and expensive workloads at the high end. A casual Plus subscriber can still think in monthly-plan terms, but a company deploying coding agents, research tools, and custom actions now has to think in meters, pools, and overruns. (help.openai.com 1) (help.openai.com 2) The result is a cleaner split inside OpenAI’s business. ChatGPT stays familiar enough to sell like software, while the most valuable agent-style work starts to bill like compute, which is usually where infrastructure companies make their money. (openai.com) (help.openai.com)