Claude Projects vs Custom GPTs: Which to Build On Now
September 30, 2026
Claude Projects vs custom GPTs looks like a like-for-like comparison. It isn't. Both let you save instructions and files so you stop re-explaining yourself, but they're built for different jobs.
- A Claude Project is a workspace. You put your documents and instructions in one place, then have many focused chats against them. It's mostly for you, or your team.
- A custom GPT is a configured assistant. You give it instructions, knowledge, and tools, then hand it to other people — including the public through ChatGPT's GPT directory.
The short answer: use a Claude Project when the work is yours and the files are big. Use a custom GPT when the point is handing an assistant to someone else. And before you build a new custom GPT, read the section below on plugins, because OpenAI's own docs now describe custom GPTs as being retired.
Quick comparison
| Claude Projects | Custom GPTs | |
|---|---|---|
| What it is | A workspace with its own chats and knowledge base | A configured version of ChatGPT for one use case |
| Built from | Project knowledge (files) and project instructions | Instructions, knowledge files, capabilities, and Actions |
| Large file sets | Paid plans switch to retrieval automatically | Knowledge files uploaded to the GPT |
| Calls outside APIs | Not part of the project definition | Yes, through GPT Actions |
| Sharing | Team and Enterprise plans, inside your organization | With an allowed audience, and publicly via the GPT directory |
| Public directory | None | GPT directory |
| Status, September 2026 | Current; a new beta version is rolling out in Claude Code | Being transitioned to plugins, per OpenAI's docs |
Custom GPTs are moving to plugins
This is the fact most comparison articles miss, and it changes the advice.
OpenAI's guide to moving custom GPT workflows to plugins says, in plain words, that it is "transitioning custom GPTs to plugins." It lays out milestones: an admin notice, a migration window, then retirement, when custom GPTs stop running and leave the GPT directory.
What moves across, according to that guide:
| Custom GPT part | After migration to a plugin |
|---|---|
| Instructions | Become a skill |
| Knowledge files | Carried over as reference files |
| Connected apps | Stay part of the plugin |
| Custom actions | Don't transfer; rebuild with a supported app or a custom MCP server |
| Selected model | Doesn't carry over |
| Conversation starters | Don't transfer one-to-one |
Two caveats keep this honest. The guide is written for ChatGPT Enterprise workspaces, and it says timelines for other plans may differ. And it doesn't publish a retirement date. So as of September 2026, custom GPTs still work — but if you're choosing where to build something you want to keep for years, you're really choosing between a Claude Project and an OpenAI plugin, not a GPT.
We covered the plugin directory and other places AI tools get distributed in our GPT Store alternatives guide, so we won't repeat it here.
Where Claude Projects are stronger
Large document sets. A Claude Project is a self-contained workspace with its own chat history and knowledge base. On paid plans, when your project knowledge approaches the context limit, Claude switches on retrieval mode automatically. Anthropic says this expands capacity by up to 10x, and you'll see Claude using a project knowledge search tool to pull the relevant parts. There's no setup. If your project knowledge later drops back under the threshold, it can return to in-context processing.
That matters when the job is "keep all 40 contracts here and answer questions across them for the next quarter."
Many chats, one context. The unit of work is the chat, not the assistant. You open a new chat for each question, and every one starts with the same files and instructions. A due-diligence project might hold a data room, with separate chats for legal risks, financial questions, and the memo draft. You never re-upload.
Team collaboration without publishing. On Team and Enterprise plans, you can share a project with specific colleagues by email, in bulk, or with the whole organization. Members get Can view (see the knowledge and instructions, chat inside it) or Can edit (change instructions and knowledge, manage members). Organization owners can switch project sharing off entirely. Nothing leaves the organization, because there's nowhere public for it to go.
It's available on every plan. Anthropic's help page says projects are available to all users, including free accounts, which are capped at five projects. Retrieval mode and sharing are the paid parts.
A newer version is arriving. Anthropic is rolling out a new version of projects, starting in Claude Code. In the Claude Code docs, a project becomes one ongoing conversation that splits work into parallel cloud threads sharing the same repositories, instructions, and memory. As of September 2026 it's a public beta on Pro and Max, not yet on Team or Enterprise, and existing chat projects keep working as before. If you live in a codebase, our guide to the CLAUDE.md file covers the other way Claude Code keeps standing instructions.
Where custom GPTs are stronger
Handing an assistant to someone who isn't you. A custom GPT is the thing itself. You configure it once, share it, and the other person opens it and starts talking. They don't need to know what files are in it or how the instructions are worded. For a client-facing FAQ bot or an onboarding helper, that shape fits better than a workspace.
Built-in integrations. This is the biggest mechanical difference. GPT Actions let a custom GPT call third-party REST APIs, with the builder describing the API schema and authentication. A user asks in plain language; the GPT works out the call. OpenAI's examples include querying a data warehouse or filing a Jira ticket. Claude's help page describes a project as knowledge plus instructions — the integration isn't part of the package.
One catch in business workspaces: OpenAI's GPT sharing controls say builders can use connected apps or custom actions in a GPT, but not both, and admins can restrict actions to approved domains.
A public directory. A GPT can be listed in ChatGPT's GPT directory, so people who'd never heard of you can find it. Claude Projects are shared inside a Team or Enterprise organization rather than listed publicly. That was the strongest argument for GPTs — and it's also the part most affected by the move to plugins: OpenAI's migration guide says that at retirement, custom GPTs "stop running and leave the GPT directory."
Tools in the box. A custom GPT can switch on ChatGPT capabilities like web search, image generation, and data analysis. Image generation in particular is something Claude doesn't do, as we covered in Claude vs ChatGPT.
The fairer match: ChatGPT Projects
If what you actually want is a Claude Project, the closest OpenAI equivalent isn't a custom GPT. It's a ChatGPT project.
According to OpenAI's Projects and chats docs, a ChatGPT project keeps related chats, uploaded files, connected sources, and project instructions together, and the instructions apply across its chats. In the ChatGPT desktop app, a local project can also attach folders on your computer, which ChatGPT can then read and change.
| Claude Project | ChatGPT project | Custom GPT | |
|---|---|---|---|
| Built for | Your ongoing work | Your ongoing work | Other people using an assistant |
| Holds | Files and instructions | Files, sources, and instructions | Instructions, knowledge, tools, Actions |
| Local folders | New Claude Code version can run a thread in a folder on your computer | Desktop local projects can attach folders | Not described in OpenAI's docs |
| Instructions apply to | Every chat in the project | Every chat in the project | Every conversation with the GPT |
So the honest three-way framing is: Claude Project vs ChatGPT project for your own work, and custom GPT (soon, plugin) for distributing an assistant.
Claude Projects vs custom GPTs: how to decide
| Your situation | Pick |
|---|---|
| Hundreds of pages you'll question for months | Claude Project |
| A team on Claude sharing one knowledge base | Claude Project |
| An assistant for clients or the public | Custom GPT today, plan for a plugin |
| It must call your own API | Custom GPT with Actions, or a plugin with an MCP server |
| You already live in ChatGPT and want a workspace | ChatGPT project |
| You're starting something new that must last | Claude Project, or build straight as an OpenAI plugin |
If you want the Claude-side way to package reusable behavior — closer to what a GPT's instructions do — look at Claude skills. Skills are also what OpenAI now converts GPT instructions into.
What neither one keeps
Both tools solve re-explaining. Neither solves re-building.
A project remembers your files, but the process you worked out in chat three still lives in chat three. A custom GPT packages a process, but it runs only inside ChatGPT, and now it's being migrated. Either way, the thing that worked is tied to one vendor's chat window.
That's the gap Taku is shaped around: mirror an AI app, skill, or workflow someone already got working into your own desktop workspace, remix it for your files, and keep it as something you run again. If you'd rather start from a setup that already works than configure one from scratch, the free app library is the place to look. Taku is in Beta, and the Mac app is available now.
Key points
- Claude Projects are workspaces: knowledge plus instructions, with many chats on top. Custom GPTs are assistants you configure and hand out.
- On paid Claude plans, projects switch to retrieval automatically as knowledge grows, for up to 10x capacity.
- Claude project sharing is Team and Enterprise only, and stays inside your organization.
- Custom GPTs win on Actions, built-in tools like image generation, and public distribution.
- OpenAI's docs say custom GPTs are being transitioned to plugins and will be retired; the Enterprise guide gives no date and says other plans may differ.
- For your own ongoing work, compare Claude Projects with ChatGPT projects instead.
FAQ
Is a Claude Project the same as a custom GPT?
No. A Claude Project is a workspace where you upload files and set instructions, then run many chats against them. A custom GPT is a configured assistant you can share with other people, with optional Actions that call outside APIs.
Are custom GPTs being discontinued?
OpenAI's developer docs say it is transitioning custom GPTs to plugins, and that at retirement custom GPTs stop running and leave the GPT directory. The migration guide targets Enterprise workspaces, gives no public date, and says timelines for other plans may differ. As of September 2026 they still work.
Can I share a Claude Project?
Yes, on Team and Enterprise plans, with specific people or your whole organization, at view or edit level. There's no public sharing or store for projects.
Which handles more documents?
On paid plans, Claude Projects move to retrieval automatically when knowledge nears the context limit, which Anthropic says expands capacity by up to 10x. For very large, long-lived document sets, that's the stronger fit.
What is the Claude equivalent of a custom GPT?
There isn't an exact one. A project covers the knowledge-and-instructions half. Skills and plugins cover the packaged, reusable-behavior half. Neither gives you a public listing like ChatGPT's GPT directory.
Can I move a custom GPT to Claude?
Treat it as a manual move. You can copy the instructions into a project's instructions and upload the same files as project knowledge. Any Actions would need rebuilding as a separate integration, which is also true when OpenAI migrates a GPT to a plugin.