OpenAI tidies enterprise rate card

OpenAI has rewritten parts of its enterprise ChatGPT pricing guide to spell out which models are gone, which remain, and what happens when users hit limits. (help.openai.com) The updated ChatGPT rate card says that, as of February 13, 2026, GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 “Instant” and “Thinking” were retired from ChatGPT. It adds that application programming interface access to those models is unchanged. (help.openai.com) For ChatGPT Business, Enterprise, and Edu customers, OpenAI says GPT-4o stayed available inside custom GPTs until April 3, 2026, even after the broader retirement in the main ChatGPT interface. Separate Business and Enterprise model-limit pages repeat the same cutoff and retirement language. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) The more practical change for buyers is the fallback rule. OpenAI’s release notes say GPT-5.4 mini is now the fallback for GPT-5.4 Thinking when paid users hit rate limits, and that the smaller model does not appear as a separate picker option. (help.openai.com 1) (help.openai.com 2) That means procurement teams and workspace admins now have a clearer answer to a basic question: when a higher-tier model is capped, what model serves the next request. OpenAI also says Enterprise customers can choose GPT-5.4 mini as the default for Auto routing, which gives admins another explicit control over model selection. (help.openai.com) OpenAI made a parallel cleanup in Codex, its coding product inside ChatGPT. The “Using Codex with your ChatGPT plan” page says Codex now supports the GPT-5.1-Codex family, with Max as the default and Mini as an optional alternative. (help.openai.com) The Codex pricing page says OpenAI changed Codex billing on April 2, 2026, for new and existing Plus and Pro users, new and existing ChatGPT Business users, and new Enterprise customers. Under that system, usage is billed by input, cached input, and output tokens, replacing per-message estimates with credits per 1 million tokens. (help.openai.com) That same page keeps the older method on the record for existing Enterprise customers and some other plans that have not yet been migrated. OpenAI says the legacy card used approximate credits per message or per pull request, and warned that actual usage could vary with task size, model choice, and reasoning requirements. (help.openai.com) The result is less mystery in two places that large customers usually press vendors on first: what model employees are actually using, and how usage turns into charges. OpenAI’s help pages now answer both questions with dated retirement notices, named fallback models, and separate Codex billing rules by plan. (help.openai.com 1) (help.openai.com 2)