From GPTs to Plugins: What Changes

The previous fourteen lessons taught you how to build a GPT: solid instructions, a knowledge base, actions connecting to APIs, and publication. OpenAI is now replacing this building block with a new object called a plugin and will retire custom GPTs on December 11, 2026. This lesson explains what a plugin is, what carries over from your GPT work, and what does not.

What Is a Plugin?

According to ChatGPT documentation, a plugin is a reusable package that helps ChatGPT accomplish a task. It can contain apps, skills, or both. Apps connect ChatGPT to other services: searching for a file in storage, updating a record. Skills describe how to do the work: summarizing a document in a consistent format, applying a reading rubric, filling out a template. OpenAI illustrates the combination of the two with a simple example: retrieving two documents via an authorized app, then comparing them against your checklist using a skill.

Plugins work in both Chat and Work modes in ChatGPT on the web, desktop, and mobile, as well as in Codex. They live in a shared directory for ChatGPT and Codex and are invoked by typing @ in the input field.

What Carries Over from Your GPT

The good news is that most of your expertise transfers directly. During migration, a GPT’s instructions become a skill, its knowledge files are copied into the plugin’s reference files, and its connected applications become apps. The model selected for the GPT, however, is not carried over. Everything you learned about writing clear instructions, structuring a knowledge base, and testing on representative cases remains directly useful.

What Does Not Carry Over

Two areas require work. First, custom actions—the API connections you built in lessons 9 through 11—are not migrated: you must rebuild them using an existing app or a custom MCP server. Second, the invocation method changes: with a GPT, the user explicitly chose the assistant; with plugins, ChatGPT can also automatically use an installed skill when its description matches the request. OpenAI therefore recommends testing both scenarios: explicit selection and a standard request.

The Timeline

The documentation is now explicit for all plans. Custom GPTs will be retired on December 11, 2026; FedRAMP workspaces and certain Enterprise workspaces with an approved exemption have until February 11, 2027. Personal accounts can no longer create new GPTs, and creation ceased on October 26, 2026, in affected Enterprise workspaces. To migrate, open My GPTs and choose Migrate to plugin when the option appears: migration starts from the last published version, the resulting plugin starts in private mode, and the original GPT becomes read-only until retirement. Your conversations with your GPTs will remain accessible afterward. Set aside time to test the plugin on your usual requests and on a difficult case.

Your Turn

Prepare a migration. Choose a GPT you created or use frequently. Note its instructions, files, connected applications, and any actions, then fill out the table above for it: what will carry over, what must be rebuilt. Keep three usual requests and one difficult case: these will be your tests after migration.

Official sources: Custom GPT retirement and migration FAQ (OpenAI Help Center: dates, migration, what carries over) · Plugins in ChatGPT · Build plugins · Moving your custom GPT workflows to plugins (ChatGPT documentation)

Key Takeaways

• A plugin groups apps (service connections) and skills (work methods), and is invoked with @
• During migration, instructions become a skill, files become reference files, and apps remain apps; actions must be rebuilt
• ChatGPT can select an installed skill on its own: test both explicit selection and standard requests
• GPT retirement on December 11, 2026 (February 11, 2027 with exemption): migrate and test before then