Custom GPTs retire from December 11, 2026, with some workspaces given until February 2027

What happens and when

ChatGPT's custom GPTs are being retired, and workspaces that rely on them now face firm deadlines. According to a dev.to migration guide published on November 10, 2026, the standard retirement date is December 11, 2026. Not every workspace shares that date: OpenAI's retirement and migration FAQ, as cited in the post, gives FedRAMP workspaces and affected Enterprise workspaces with a qualifying, approved deferral until February 11, 2027.

The later date is not a blanket extension. The guide stresses checking the notice attached to the account or workspace where the GPT was created, not merely the account used to open it. Migration availability also varies, so an announced target date is no guarantee that the migration option has appeared for a given account. For affected Enterprise workspaces there is a separate creation cutoff of October 26, 2026, from which FedRAMP workspaces are excluded. Existing, unmigrated GPTs remain editable until their applicable retirement date.

At retirement the GPTs and their pages become inaccessible, but OpenAI's FAQ says existing conversations with custom GPTs will remain accessible. The migration guide adds that migration will still be possible after retirement, although the GPT itself will no longer run — a recovery path, not a reason to postpone.

What transfers and what does not

Migration converts a GPT into a plugin, a package that can bundle skills and connected apps. Per the dev.to breakdown of OpenAI's guidance:

• Instructions become a skill inside the plugin.
• Knowledge files are copied into plugin reference files, though their actual use still needs checking.
• Connected apps are added as apps, but users still need the relevant access and authorization.
• Custom actions do not transfer and must be rebuilt separately.
• The selected model does not carry over; Enterprise workspaces get enterprise defaults.
• Conversation starters do not transfer one-to-one.
• Capability settings do not guarantee the same tools or behavior.
• Existing sharing does not carry across; a replacement plugin starts private.

OpenAI's FAQ is blunt about expectations: "A migrated plugin may respond differently."

Creators and users face different jobs

The guide separates builders from consumers. Creators should finalize the GPT first, because migration uses the latest published version and drafts or unpublished edits do not transfer; publishing does not require making the GPT public. Once migration runs, the original becomes read-only and — per the FAQ — its creator cannot delete it, so future changes happen in the plugin.

Users who did not build a GPT are in a weaker position: permission to use someone else's GPT does not confer permission to migrate it. In Enterprise workspaces, an admin can perform the migration when the required access is enabled. The guide also warns that being able to open the old GPT is not proof of access to its replacement, which may need separate installation and entitlements.

Testing before you rely on the replacement

OpenAI recommends running a few familiar prompts plus at least one harder case, and its FAQ asks migrators to verify four things: the right skill is selected and follows instructions, the expected reference material is used, answers or files are complete and in the required format, and the necessary tools and integrations are available. The dev.to guide adds testing both explicit skill selection and an ordinary task request, because automatic selection is task-dependent — an installed plugin will not necessarily run for every prompt.

The author also advises data hygiene when comparing outputs: remove or anonymize personal and confidential material, exclude secrets, passwords and API keys, and treat text comparison as an aid rather than proof that integrations work.

One caveat worth flagging: the dev.to author discloses that the post was drafted and edited with AI tools and that they have not personally performed a GPT-to-plugin migration. The deadlines and mechanics come from OpenAI's FAQ and migration guide; the preparation checklist and rehearsal scenarios are the author's own.

Why it matters

Custom GPTs became a lightweight way for teams to package instructions, files and tools without writing code. Their retirement pushes that functionality into a plugin model with different boundaries: custom actions and model choice disappear, sharing must be re-established, and behavior may shift even when instructions and files carry over. Any workspace with GPTs embedded in real processes should inventory them now, check the notice on the workspace that created them, and budget time to rebuild integrations and retest outputs before December 11, 2026 — or February 11, 2027 where a deferral applies.