ChatGPT and Claude both have projects, and from a distance they look like the same feature: a folder with some files, a few instructions and the chats that belong together. Look closer and they disagree on two things that matter a lot in practice: what a project remembers, and who you can bring into it.
Below they sit side by side. We didn't pick a winner first and work backwards. Everything about Claude was checked on Anthropic's help center and everything about ChatGPT on OpenAI's help center on 18 November 2026. Both companies change limits and plans often, so check the numbers again before you build a workflow on them.
What a project is in ChatGPT and in Claude
The basic idea is the same in both apps. A project groups chats, files and instructions, and every chat you start inside it begins with that material.
In Claude, Anthropic calls projects "self-contained workspaces with their own chat histories and knowledge bases". You set project instructions, which Claude uses in every chat of the project, and you upload project knowledge: documents, text, code. You can also move an existing chat into a project.
ChatGPT's version holds the same three things: chats, files and instructions, with the instructions applying only inside that project.
You get projects on every plan in both apps, free ones included. Where they split first is size.
ChatGPT and Claude projects at a glance
| ChatGPT | Claude | |
|---|---|---|
| Chats, files, instructions | ||
| Free plan | yes, up to five projects | |
| Memory per project | default or project-only | its own, plus account memory |
| Sharing | every plan | Team and Enterprise |
Files and limits
ChatGPT caps the number of files per project, and the cap depends on the plan: 5 files on Free, 25 on Go and Plus, 40 on Pro, Business and Enterprise. If you hit it, OpenAI suggests removing older files, combining them, or splitting the work across more projects.
Files per ChatGPT project, by plan
Claude sets its limits elsewhere. Free users can create at most five projects. On paid plans (Pro, Max, Ultra, Team and Enterprise), when project knowledge gets close to the context window, Claude switches on RAG by itself: it searches the knowledge and pulls in the relevant parts, which according to Anthropic expands what a project can hold by up to ten times. No setup is needed.
In ChatGPT, then, you count files. In Claude you mostly count text. For a project built around a thick manual or a stack of contracts, that's the line to read twice.
Memory: what each project remembers
Memory is where the two apps drift furthest apart, and it's easy to misread.
Start with Claude. Files and instructions reach every chat in the project, but the chats themselves don't share context. Anthropic says it plainly: "Context is not shared across chats within a project unless the information is added into the project knowledge base." What fills the gap is memory. Each project has its own, and inside a project Claude uses that project memory plus your account memory from chats outside projects, never the memory of another project. To seal a project off completely, you turn off "Use account memory" on the project page. On paid plans you can also ask Claude to search past chats, and inside a project that search stays within the project.
ChatGPT hands you a choice instead, one project at a time. With default memory, the project can use your saved memories, and what it can reach outside the project depends on your plan. With project-only memory, chats can refer to other conversations in the same project, but not to your saved memories or to conversations outside it. You can switch an unshared project between the two in Project settings, and OpenAI warns the change can take a few hours.
One detail catches people out: a shared ChatGPT project always uses project-only memory and can't reach a member's personal memories.
The goal is the same in both, keeping one client or topic from bleeding into another. Claude uses a memory per project plus a switch; ChatGPT uses a mode you pick.
Sharing a project with other people
On sharing, the two are far apart.
Claude offers project sharing on Team and Enterprise plans, with other members of your organization. You add people by name or email and give them "Can view", to read the knowledge and chat in the project, or "Can edit", to change instructions and knowledge. Someone outside your organization can't be added.
OpenAI's help center lists project sharing on every ChatGPT plan, from Free to Enterprise. On personal plans you invite individual people; Business, Enterprise and Edu can also invite workspace groups, and in a managed workspace invitations may be limited to its members. Each person gets chat access or edit access, and edit access lets them change instructions, upload or remove files and invite others. How many people can join follows the owner's plan: 5 on Free, 10 on Go and Plus, 100 on Pro.
Collaborators on a shared ChatGPT project
So if you work with freelancers or clients outside your company, a personal ChatGPT plan can bring them in, while Claude can't. The catch: whoever you invite has to work inside ChatGPT. We looked at what that means for an agency with many clients in one AI context per client.
Where both kinds of project stop
For all their differences, both hit the same wall. A project lives inside one product.
A Claude project doesn't exist in ChatGPT, and a ChatGPT project doesn't exist in Claude. If half the team uses one app and half the other, each topic needs two projects, kept in sync by hand. Claude Code and Codex have their own instruction files on top of that. Anthropic is rolling out a new kind of project, starting with Claude Code, but it's still a Claude feature.
And knowledge is whatever someone uploads. When the brief changes, someone replaces the file. A decision made in a chat stays in that chat, or ends up in a memory the app manages, not in a document the whole team can open and correct.
Where a project's context stops
If everyone you work with uses the same app, that wall may never bother you, and projects are the quickest way to keep a topic in order. We wrote about when Claude Projects are enough and when they aren't.
How Cobrain fits
Cobrain sits outside both apps. Your notes are Markdown files in folders, behind a remote MCP address you connect to ChatGPT, Claude, Claude Code, Codex and Cursor. Each AI reads and writes the same notes, so a decision made in Claude is there when ChatGPT drafts the email.
You can keep using projects for instructions and reference files, and put the facts that change in shared notes. That's the idea behind one memory across AIs. The notes can travel to people too: you share a branch with any email address, to read or, on Pro, to write, and whoever receives it never pays. To start with ChatGPT, the ChatGPT guide takes a few minutes.
Quick answers
What is the difference between ChatGPT Projects and Claude Projects? Both group chats, files and instructions. They differ on memory (ChatGPT lets you choose default or project-only memory, Claude gives each project its own memory plus account memory) and on sharing (every ChatGPT plan, only Team and Enterprise in Claude).
Can I share a Claude Project with someone outside my company? No. Claude project sharing is on Team and Enterprise plans, with members of your organization. ChatGPT allows project sharing on every plan, though a managed workspace may limit invitations to its members.
How many files can a ChatGPT project hold? According to OpenAI's help center, 5 on Free, 25 on Go and Plus, and 40 on Pro, Business and Enterprise. Limits change, so check the current page.
Can ChatGPT use my Claude Project, or the other way round? No. Each project stays in the app that created it. To give both AIs the same context, keep it in notes they can both reach, for example through an MCP server.



