OpenAI is retiring custom GPTs in Enterprise workspaces: three dates you need to know
OpenAI has set three hard dates for custom GPT retirement in Enterprise workspaces, ending with full retirement on 11 December 2026.
Custom GPTs, the build-your-own ChatGPT assistants OpenAI introduced in November 2023, are being retired. If your organisation runs ChatGPT Enterprise and has GPTs in its workspace, OpenAI has now published three specific dates that mark the end of the road, and the window to act is tighter than it might look.
The three dates for Enterprise workspaces
OpenAI’s Enterprise and Edu release notes set out the following timeline:
- 17 September 2026, migration tooling becomes available, letting GPT creators begin moving their assistants to the new plugins system.
- 25 September 2026, creation of new custom GPTs in Enterprise workspaces ends. Existing GPTs remain usable, but you cannot build new ones.
- 11 December 2026, all existing custom GPTs retire. After this date, they leave the GPT directory and users must be on the new plugins system.
OpenAI states explicitly that all dates are subject to change, so it is worth watching your workspace’s in-product notices for any updates. Admins received a preparatory notice on 11 September 2026 to help organisations get ready.
One important distinction if you use multiple ChatGPT plans: the 25 September creation cutoff applies to Enterprise workspaces specifically. For other plans, the official migration FAQ lists the end-of-creation date as 26 October 2026. The December retirement applies across the board.
Why is this happening?
OpenAI replaced its App Directory with a Plugin Directory on 9 July 2026, when it launched ChatGPT Work. Plugins are the successor format, and they are designed to do more than custom GPTs could: they combine reusable instructions, reference files, and connected apps into a single package that works across Chat, Work, and Codex.
The direction of travel had been visible for a while. On 22 April 2026, OpenAI launched Workspace Agents, Codex-powered agents that connect to tools such as Slack, Salesforce, and Gmail. Custom GPTs, meanwhile, had been in routine maintenance mode. The retirement formalises what was already the practical reality.
What carries over when you migrate, and what does not
This is the part that requires the most attention. OpenAI’s migration workflow is not a one-click clone.
What does carry over:
- Your GPT’s instructions become a skill in the new plugin.
- Connected apps transfer across.
What does not carry over:
- Custom actions must be rebuilt, either with a supported connector or a custom MCP server.
- The model selected for the GPT does not transfer to the replacement plugin.
Only the GPT’s creator or a workspace admin can migrate it, and the GPT must be published before migration is possible. Public sharing is not required, but it does need to be published.
Once you migrate, the original GPT stays usable until 11 December 2026 but becomes read-only. You cannot make further edits to it, and the creator can no longer delete it. Any changes you want to preserve need to be made before you trigger the migration. Migrated links should redirect to the replacement plugin for users who have access.
What this means for you
The practical steps depend on your relationship to any given GPT.
If you created GPTs in an Enterprise workspace, finish any outstanding edits and publish them before 17 September 2026. When migration tooling arrives, work through each GPT methodically: save a few familiar prompts so you can compare results, check reference files, and pay extra attention to any that use custom actions. Those integrations will need to be rebuilt, so allow more time than you think you need and test the replacement before the December deadline.
If you use GPTs that someone else created, check who built them and whether that person is aware of the timeline. If a GPT belongs to a colleague who has left or is not paying attention to admin notices, it may simply disappear in December without a replacement. Identify your dependencies now rather than in November.
If you manage a workspace, the admin controls for agent building, publishing, and Slack usage are available now. ChatGPT workspace agents are off by default and need to be enabled for eligible workspaces. Note that workspace agents are not available for Enterprise workspaces with EKM at launch. Ensure the ChatGPT Desktop app is on version 26.812.10818 or later for new admin controls to take effect.
If you use GPTs across multiple accounts or workspaces, the timeline and permissions that apply are those of the workspace where the GPT was created, not where you access it from. Check the source workspace for each GPT you rely on.
A tight window
Eight days separate the arrival of migration tooling on 17 September and the end of new GPT creation on 25 September. For large organisations with many GPTs, that is a short runway to triage what needs migrating, identify who owns each one, and begin the process. The sensible approach is to audit your GPT inventory now, before tooling even arrives, so you are not making decisions under pressure.
The 11 December retirement date gives a reasonable amount of time to complete migrations and test replacements, but only if preparation starts well before September.
OpenAI’s migration FAQ is the best single reference for the mechanics of what transfers and how. Keep an eye on your workspace notices for any date changes, and treat the December deadline as fixed until told otherwise.