OpenAI shifts to enterprise fit

OpenAI is no longer just selling a smarter chatbot seat. In April 2026, its help pages started spelling out something closer to business software: different seat types, token-based usage, app actions, and custom tools that can be wired into company workflows. (help.openai.com, help.openai.com) The clearest change is that ChatGPT Business and ChatGPT Enterprise now have two seat types: a standard ChatGPT seat and a Codex-only seat. That is the kind of menu companies expect from software vendors when they are deciding who gets full access and who just needs one narrow job done. (help.openai.com, help.openai.com) Codex is OpenAI’s coding agent, and OpenAI says it can work in the cloud or inside local tools like a terminal and integrated development environment, which is the app programmers use to write code. A Codex-only seat means a company can buy the coding worker without buying the full general-purpose assistant for every employee. (help.openai.com, help.openai.com) At the same time, OpenAI changed how some of that work is billed. On April 2, 2026, the Codex rate card moved new ChatGPT Business and new ChatGPT Enterprise plans from per-message pricing to token-based pricing, which means companies are charged more like they are in the application programming interface business, where usage is metered by how much text goes in and out. (help.openai.com, help.openai.com) That pricing shift sits next to a product shift inside GPTs, which are custom versions of ChatGPT built for a task or team. OpenAI’s release notes say Plus, Pro, and Team users got broader model choice for Custom GPTs earlier, and Enterprise and Education users can now choose from the full set of ChatGPT models when building them. (help.openai.com, help.openai.com) A Custom GPT stops being just a prompt template once it gets actions. OpenAI’s GPT documentation says builders can configure capabilities, apps, and actions, and its business GPT help page says custom actions work with all models except the pro-series models, which turns a GPT from “answer my question” into “go do this job in another system.” (help.openai.com, help.openai.com) OpenAI has also been adding more of those system connections. On March 27, 2026, Business and Enterprise release notes said Box, Notion, Linear, and Dropbox app updates added new actions, including write capabilities where supported, so the model can do more than read a file and summarize it. (help.openai.com, help.openai.com) This is what enterprise software buyers usually ask for after the demo phase. The first question is whether the model can do the task at all, but the next questions are who can use it, what it can touch, how usage is billed, and whether it can take an action inside Box, Notion, Linear, Dropbox, or code tools without a person copying and pasting between tabs. (help.openai.com, help.openai.com, help.openai.com) Even the public pricing page now reads more like a business catalog than a model launch page. OpenAI lists Plus, Pro, Business, and Enterprise as paid plans priced per user per month, while the help center adds flexible credits and usage-based Codex seats on top, which is a more formal commercial structure than “pick the best model and start chatting.” (openai.com, help.openai.com) There is one wrinkle in the timing: several help pages now say GPT-4o and GPT-4.1 have been retired from ChatGPT as of February 13, 2026, with GPT-4o briefly retained inside some Custom GPT setups until April 3, 2026. That means the story is not really about one named model anymore; it is about the packaging around models, the actions attached to them, and the billing rules that decide where they fit inside a company. (help.openai.com, help.openai.com, help.openai.com)