OpenAI Shifts Pricing Model

OpenAI changed one of the quietest but most important parts of its business on April 2, 2026: how it charges companies for coding work inside ChatGPT. Codex, its software-writing tool, is no longer billed by message for new ChatGPT Business plans and new ChatGPT Enterprise plans; it now uses credits tied to token usage, which is closer to how cloud computing bills for electricity than how texting bills for messages. (help.openai.com) A message-based price treats a one-line bug fix and a 20-minute code review as the same unit. A token-based price breaks usage into the actual chunks of text the model reads and writes, so a short prompt with a short answer costs less than a long prompt with a long answer. (help.openai.com) (developers.openai.com) OpenAI says the April 2 change applies to new and existing ChatGPT Business customers and to new ChatGPT Enterprise customers. Existing ChatGPT Plus, ChatGPT Pro, and older Enterprise and education customers stay on the legacy Codex rate card for now, with migration planned in the coming weeks. (help.openai.com) The reason this is bigger than a billing tweak is that token pricing makes ChatGPT look more like the OpenAI application programming interface, which is the company’s developer service sold by usage. The same company can now sell a seat for access and then sell extra credits when heavy users burn through shared limits. (help.openai.com) (openai.com) OpenAI’s help pages describe that system in plain terms: Business users get per-seat limits for advanced features, and if they cross those limits, a workspace that bought credits can keep them going from a shared pool. That turns a flat software subscription into something closer to a mobile phone plan with extra data once the included allowance runs out. (help.openai.com) The timing lines up with a shift in where OpenAI says its money is coming from. On April 8, 2026, CNBC reported that OpenAI chief revenue officer Denise Dresser said enterprise now accounts for about 40% of company revenue and could reach the same size as consumer revenue by the end of 2026. (cnbc.com) That 40% figure changes how to read the pricing move. If business customers are becoming nearly half the company, then measuring expensive workloads by tokens instead of by message gives OpenAI a cleaner way to map heavy usage to revenue, especially for coding agents that can generate long outputs and inspect large codebases. (cnbc.com) (help.openai.com) It also solves a basic mismatch in artificial intelligence coding work. A coding agent does not behave like a chat window where each turn is roughly similar; one task may read thousands of lines of code, call tools, and write tests, while another task may only rename one variable, so charging per message can hide the true cost of the first job. (help.openai.com) (developers.openai.com) OpenAI has already been moving more of ChatGPT toward this hybrid model. Its Business and Enterprise pricing pages still sell seats per user per month, but the help center now layers credits on top for advanced tools like Deep Research, image generation, voice, thinking models, and Codex. (openai.com) (help.openai.com) So the product that millions of people know as a chatbot is being rebuilt underneath as metered infrastructure for companies. The seat gets a team in the door, and the token meter starts running when that team uses the expensive parts hard enough. (openai.com) (help.openai.com)