Claude Skills 與自訂 GPT：哪個適合你？

Yes. OpenAI is replacing them with Plugins. Creation of new custom GPTs ends on 25 September 2026, and existing custom GPTs are scheduled to stop running and leave the GPT directory on 11 December 2026. OpenAI publishes the timeline against Enterprise workspaces with other plans expected to follow, and notes dates can vary by workspace, so check its retirement FAQ for your own account.

Not on a personal account. OpenAI states that new GPT creation and publishing are not available on personal ChatGPT accounts, including Free, Go, Plus and Pro. Only Business, Enterprise and Edu workspaces can still create one, and that ends on 25 September 2026 as well. You can still use GPTs other people made until the retirement date.

They keep working until the retirement date. Once you migrate one it becomes read-only, so make any edits you still want before migrating rather than after. On 11 December 2026 they stop running, and migrated links are planned to redirect to their replacements. Copy the instructions and download the knowledge files before then, because nobody hands those back afterwards.

No. OpenAI is explicit that custom actions do not transfer through the migration workflow, so any API integration has to be rebuilt by hand. This is the part of a migration people underestimate. Write down what each GPT connected to before you start.

A custom GPT is a configuration stored on OpenAI's servers: instructions, up to 20 knowledge files of up to 512 MB each, switched-on capabilities, and optional actions calling external APIs. A Claude skill is a folder on a filesystem with a SKILL.md at its root, which you own and can read. The GPT can call your systems and host your documents; the skill file cannot. The skill file cannot be retired by a vendor; the GPT is being retired now.

The instructions yes, the runtime no. A SKILL.md is plain Markdown, so you can paste its contents into ChatGPT custom instructions, a Gemini gem or Copilot and the method travels intact. What does not travel is Claude's Skills system: automatic triggering when a request matches, staged loading, and bundled scripts. Elsewhere the instructions simply apply to the whole chat.

To upload one in the Claude app, yes. Anthropic makes Skills available on Pro, Max, Team and Enterprise, with code execution enabled under Settings then Capabilities. On the free plan there is no Skills menu, but the file still works: paste the SKILL.md contents into the conversation or project instructions, losing automatic triggering but keeping the method.

A skill file, and it converts almost line for line. Most custom GPTs were never integrations - they were a description of a role, a method, a voice and a standard for finished work, which is exactly what a SKILL.md holds. If your GPT genuinely called external systems, your route is Plugins instead; if it was mainly a container for documents and history, a ChatGPT Project is the closer match.

Treat it like installing software. Anthropic advises using skills only from sources you trust and auditing anything from an unknown source first, because a skill can direct Claude to run code and use tools in ways its description does not mention. The advantage of the format is that auditing is actually possible: open the SKILL.md and any bundled scripts and read them. You could never inspect a hosted GPT that way.