OpenAI Schedules Three GPT-5 API Models for Shutdown on April 1, 2027

OpenAI is giving developers six months to move off GPT-5.3-Codex, GPT-5.1 and GPT-5.4-Nano—not indefinite access to keep existing applications unchanged. An October 1, 2026 deprecation notice schedules all three models for removal from the API on April 1, 2027. The deadline turns OpenAI’s general retirement policy into a concrete migration task for software that relies on these models.

The announcement names a replacement for each retiring model. OpenAI recommends GPT-6 Sol for both GPT-5.3-Codex and GPT-5.1, and GPT-6 Luna for GPT-5.4-Nano. Developers are instructed to complete the move before the shutdown date, rather than wait until access disappears.

OpenAI describes the notice period as time to evaluate replacement models, test how applications behave and finish migrations. Its stated process therefore includes checking the replacement in an application, not merely identifying a new model name. The six-month interval is intended to leave room for that work before the old models become unavailable.

OpenAI’s recommended migration paths

01gpt-5.3-codex

Recommended replacement

gpt-6-sol

OpenAI lists gpt-6-sol as the recommended replacement for gpt-5.3-codex.

02gpt-5.1

Recommended replacement

gpt-6-sol

OpenAI lists gpt-6-sol as the recommended replacement for gpt-5.1.

03gpt-5.4-nano

Recommended replacement

gpt-6-luna

OpenAI lists gpt-6-luna as the recommended replacement for gpt-5.4-nano.

Deprecated now, inaccessible at shutdown

The terminology matters. OpenAI considers a model deprecated as soon as it announces retirement. That status begins the retirement process; it does not mean the model has already been switched off. Shutdown is the point when the model or endpoint is no longer accessible. OpenAI uses “sunset” and “shut down” to mean the same thing.

“Legacy” is a different designation: it means a model or endpoint no longer receives updates. OpenAI uses that label to signal where its platform is heading and says legacy offerings should be expected to face deprecation eventually. This notice goes further than a legacy label because it supplies a specific end date.

Six months is not the rule for every model

All three models in this announcement receive six months’ notice. OpenAI’s broader model retirement policy sets different minimum periods depending on the type of model. It explicitly includes Codex models among specialized variants, a category with a shorter minimum than the window granted in this announcement.

• Generally available models: at least six months’ notice before retirement.
• Specialized variants, including Codex, chat and deep-research variants: at least three months.
• Preview models: potentially much shorter notice, such as two weeks.

Those minimums have an exception. OpenAI says safety or compliance concerns can require a faster retirement, in which case it will provide as much notice as reasonably possible. It also discourages using preview models for business-critical production work unless the customer can migrate on short notice. A preview’s shorter window is therefore a stated deployment constraint, not the timetable applied to these three retirements.

Notification and a conditional route to longer access

OpenAI says it emails customers actively using a model when it announces deprecation and records the change in its documentation. Its general process also includes blog posts for larger changes. The documentation is where developers can find the scheduled shutdowns and the company’s recommended replacements.

There is a possible exception to losing access at a model’s shutdown date, but it is conditional. OpenAI says developers may, in some cases, provision dedicated capacity for continued access and directs them to its sales team to explore that option. It does not present dedicated capacity as an automatic extension for everyone using a deprecated model.