OpenAI Is Retiring Custom GPTs on Dec 11 — Migrate to Plugins or Lose Them
OpenAI is killing custom GPTs on December 11. Custom actions will break, and you must migrate to plugins manually.
Tools · Source: Campus Technology
What happened
OpenAI announced it will retire custom GPTs on December 11, 2026. The company is shifting all customized assistant workflows into a unified plugin architecture across ChatGPT and the Codex coding agent. Personal account users on Free, Go, Plus, and Pro tiers can no longer create or publish new custom GPTs. For affected Enterprise workspaces, creation of new custom GPTs shuts down on October 26, though accounts with approved deferrals can keep running existing setups until February 11, 2027.
OpenAI is replacing dedicated assistant bots with plugins that package reusable skills, reference files, and connected external apps. Existing creators can trigger a migration flow inside the My GPTs interface to convert their bots into plugins. When migrated, a GPT's instructions convert into a skill, and uploaded knowledge files become plugin reference files. However, the original GPT becomes read-only, and OpenAI warns that migrated workflows may respond differently and must be thoroughly retested.
The migration process has severe gaps that will break complex production workflows. Existing chat histories do not carry over to the new format, and the selected base model does not transfer. Most critically, custom actions that connect GPTs to external APIs do not migrate at all. Any builder relying on API integrations must manually rebuild them using supported connected apps or a custom Model Context Protocol (MCP) server.
Key facts
- Dec 11, 2026 — Scheduled retirement date for custom GPTs across OpenAI accounts
- Oct 26, 2026 — Cutoff date for creating new custom GPTs in affected Enterprise workspaces
- Feb 11, 2027 — Extended deadline for Enterprise customers with approved deferrals
- Zero — Number of custom actions or conversation histories that transfer automatically during migration
Why it matters
This deprecation is a massive disruption for anyone who built products, internal business automations, or customer-facing assistants on the custom GPT framework. Sharing settings do not transfer, meaning public or shared GPT links will stop working once retirement hits. Migrated plugins default to private, requiring teams to re-invite users and reconfigure workspace access permissions from scratch.
The change reveals OpenAI's broader platform strategy: retiring isolated toy chatbots in favor of modular, enterprise-ready tools across ChatGPT and Codex. Shifting toward MCP servers and reusable skills forces builders to use standardized developer protocols. If you treated custom GPTs as a stable, long-term foundation for your AI startup or client deliverables, you now have a hard deadline to re-engineer your stack.
For builders
Rebuild custom actions with MCP servers
If your custom GPT calls external APIs through custom actions, migration will break it instantly. You must either rewire your endpoints into an approved connected app or spin up a dedicated MCP server before December 11. Teams that wait will see their automations fail silently when the cutoff passes.
Audit and retest prompt logic
Migrated plugins do not preserve original model selections or execution environments, and OpenAI warns output behavior can change. Builders need to evaluate whether reference documentation and skill instructions still format outputs correctly. Run side-by-side prompt benchmarks before pushing the plugin to internal teams or paying clients.
Reconfigure client distribution and permissions
Public links to custom GPTs will become inaccessible after December 11, cutting off external users and clients. Migrated plugins launch as private, and access depends strictly on workspace permissions set by administrators. Agencies and software consultants must reset their client delivery pipelines and verify workspace-level plugin sharing rules immediately.
My take
If you built a business model entirely inside OpenAI's walled playground, you just learned the platform risk lesson the hard way. Custom GPTs were an experimental sandbox, and killing custom actions proves OpenAI will break your workflow whenever it wants to standardize infrastructure. Stop treating closed web interfaces like backend infrastructure and build on real protocols.
Original reporting: Campus Technology. This is my rewrite and opinion.