Custom GPTs vs Claude Skills

What differs, what OpenAI’s late-2026 GPT retirement means, and how to write instructions that survive a move.

This comparison is current as of 6 November 2026 and is based on Anthropic’s Agent Skills documentation and OpenAI’s public help articles, including the Custom GPT retirement and migration FAQ. OpenAI’s dates and product names can change; trust the notice inside your own ChatGPT account if it differs.

In this guide

The short version
Where Custom GPTs stand in late 2026
Side-by-side
Which should you use?
If you already have GPTs

The short version

A Claude Skill is a portable folder. Claude reads its name and description, and when your request matches, it loads SKILL.md (and only then any linked files or scripts). The same Skill idea works across Claude products, but you install it separately on each surface: Claude Code, claude.ai, or the API.

A Custom GPT is a configured ChatGPT: instructions, optional knowledge files, conversation starters, and optional Actions, living inside ChatGPT. OpenAI has said GPTs are not a way to embed ChatGPT in an external website, and that a GPT does not use your personal memory or custom instructions the way the main ChatGPT profile does.

Those two designs were already different. In 2026 they diverged further: OpenAI is retiring Custom GPTs and moving creators toward plugins, where a GPT’s instructions become a skill inside the plugin and its knowledge files become reference files.

Where Custom GPTs stand in late 2026

According to OpenAI’s retirement FAQ and the Creating and editing GPTs help article:

Custom GPTs are scheduled to retire on 11 December 2026.
FedRAMP workspaces, and Enterprise workspaces with a qualifying approved deferral, have a retirement date of 11 February 2027. That later date does not apply to every workspace.
Existing GPTs remain usable until the retirement date that applies to them. After you migrate a GPT, the original stays usable until retirement but becomes read-only.
At retirement, custom GPTs and their GPT pages become inaccessible. OpenAI says existing conversations with custom GPTs remain accessible after retirement.
For affected Enterprise workspaces, new GPT creation ended on 26 October 2026, except FedRAMP workspaces. OpenAI has also stated that new GPT creation and publishing are not available on personal ChatGPT accounts (and that publishing new GPTs is not available there).
When migration is available, creators use Migrate to plugin from My GPTs. Migration uses the latest published version; drafts do not transfer. Publishing does not require public sharing.
What transfers: instructions become a skill in the plugin; connected apps become plugin apps; knowledge files are copied as reference files. The selected model does not carry over. Custom actions do not transfer automatically and may need a rebuilt integration, sometimes via a custom MCP server.
A migrated plugin starts private. Users of the old GPT do not automatically get the replacement.

OpenAI also notes that ChatGPT may use a relevant installed skill automatically when its description matches a request — a discovery idea that will sound familiar if you already write Claude Skills.

None of this means ChatGPT instruction-writing is obsolete. It means the packaging is changing: from “a GPT you open by name” toward reusable skills inside plugins, plus whatever custom instructions your ChatGPT account still supports.

Side-by-side

	Claude Skills / Agent Skills	ChatGPT Custom GPTs	ChatGPT plugins, after migration
What it is	A folder with `SKILL.md`, optional scripts, references, and assets	A configured ChatGPT with instructions, knowledge files, and optional Actions	A plugin that can bundle skills (instructions), reference files, and connected apps
Where it runs	Claude Code, claude.ai, Claude API and related cloud platforms — installed per surface	Inside ChatGPT only, until retirement	Inside ChatGPT, where plugins are available to that account
How it activates	Automatically when the task matches the description; in chat you can also pick a Skill with `/` on claude.ai	You open that GPT or @-mention it	You select a plugin or skill; ChatGPT may also auto-select a matching installed skill
Knowledge	Bundled files loaded on demand; unread files do not consume context	OpenAI documented up to 20 knowledge files per GPT, each up to 512 MB, while GPTs were the product	Migrated knowledge files become plugin reference files
Code and tools	Optional scripts; runtime and network rules depend on the Claude product	Built-in capabilities plus optional Actions (OpenAI noted a GPT could use apps/actions with important limits)	Connected apps migrate; custom actions do not and must be rebuilt
Portability	The folder is a file you can zip, put in git, or share. The `SKILL.md` format is documented publicly at agentskills.io	Lives in ChatGPT’s GPT configuration	Lives in ChatGPT’s plugin system; treat official migration docs as the source of truth for format details
Status in Nov 2026	Actively documented and expanding across Claude products	Scheduled to retire on 11 December 2026 for most workspaces	OpenAI’s stated replacement path for GPT creators

Which should you use?

Choose a Claude Skill if you work in Claude Code or want a version-controlled folder you can copy between projects, attach in the API, or teach teammates to audit. It is the better fit for procedures that benefit from scripts and on-demand reference files.
Do not start a new Custom GPT on an account that can no longer create them, and do not plan a product that depends on GPT pages existing after the retirement date in your notice.
If you still maintain an existing GPT, export or copy its instructions and knowledge now, finish essential edits, and migrate to a plugin when the button is available. Save a few evaluation prompts so you can compare outputs.
If you write instructions today, write them so they survive a move: clear role, workflow, output format, and “when to use this” language. That shape maps onto a Claude SKILL.md and, per OpenAI, onto a plugin skill as well.

Skillory publishes both Claude SKILL.md files and ChatGPT-style instruction sets for that reason. The instruction sets in our library can be pasted into an existing GPT you can still edit, into a plugin skill, or adapted into a SKILL.md with a proper description.

A sane migration checklist

Copy the GPT’s instructions and file list into a document you control.
Note any custom actions — those need a separate rebuild.
Publish the GPT if you intend to use OpenAI’s migrator; unpublished drafts do not transfer.
Migrate, then test with familiar prompts and one harder case, as OpenAI recommends.
Share the new plugin separately; old users are not added automatically.
If you also use Claude, turn the same procedure into a Skill folder so the workflow is not trapped in one vendor’s UI.