OpenAI model churn continues
OpenAI is rolling out frequent model and pricing changes and plans to retire older Codex models starting April 14, which is creating a moving target for teams that rely on specific model SKUs or billing assumptions. The churn is prompting the recommendation—seen in industry commentary—that AI be designed as a replaceable component inside governed workflows rather than hard-wired into critical processes. (help.openai.com, thetechoutlook.com)
OpenAI is changing the plumbing while companies are still standing on it. On April 14, 2026, six older Codex models are scheduled to disappear for people using Codex through a ChatGPT account, including `gpt-5.2-codex`, `gpt-5.1-codex-mini`, `gpt-5.1-codex-max`, `gpt-5.1-codex`, `gpt-5.1`, and `gpt-5`. (github.com) That matters because a model name is not just a label. In many teams, that exact name is wired into scripts, quality tests, approval rules, and cost forecasts, so removing one model can break more than one workflow at once. (help.openai.com) OpenAI is also splitting the product in two different ways at the same time. The company says model availability in ChatGPT is separate from Codex, and separate again from the application programming interface, which is the paid developer service for building software on top of OpenAI models. (help.openai.com) That means one model can vanish in one place and keep running in another. OpenAI retired several older models from ChatGPT on February 13, 2026, while saying there were no application programming interface changes tied to that ChatGPT retirement. (openai.com, help.openai.com) Pricing is moving too. OpenAI says that as of April 2, 2026, ChatGPT Business changed its seat structure, cut standard seat prices by 5 United States dollars per month, and added a Codex-only seat that is billed on usage instead of a flat subscription. (help.openai.com, help.openai.com, help.openai.com) So the moving parts are now stacked on top of each other. A team can keep the same coding workflow, but still see different behavior because the model changed, the seat type changed, or the billing path changed. (help.openai.com, help.openai.com) OpenAI’s own product pages show where the company wants users to go next. The current Codex models page tells developers to start most tasks with `gpt-5.4`, use `gpt-5.4-mini` for cheaper and faster work, and notes that `gpt-5.3-codex-spark` is a research preview for ChatGPT Pro. (developers.openai.com) The pattern is not one big shutdown. It is a steady replacement cycle: ChatGPT model retirements in February, GPT-5.1 removed from ChatGPT on March 11, 2026, a new `gpt-5-codex-mini` option added on March 31, 2026, and another Codex cleanup set for April 14, 2026. (help.openai.com, developers.openai.com, github.com) That is why more teams are treating the model like a replaceable engine instead of a permanent foundation. If the workflow says “run a coding model, check the diff, run tests, require human approval,” swapping `gpt-5.1-codex` for `gpt-5.4` is annoying but survivable; if the workflow says “this one stock-keeping unit is the workflow,” every retirement becomes a fire drill. (developers.openai.com, help.openai.com) The practical takeaway is boring in the way seat belts are boring. Teams that depend on OpenAI now need a model registry, fallback choices, cost alerts, and a date calendar for deprecations, because the old assumption that one model name will sit still for a year no longer matches the product. (developers.openai.com, help.openai.com, help.openai.com)
