Claude Artifacts vs ChatGPT Canvas

Comparing Claude Artifacts with ChatGPT Canvas today means comparing Artifacts with what replaced Canvas. OpenAI’s help centre says canvas is not available in its current chat models (GPT-5.5 Instant and Thinking when this was written), and that writing and coding now happen in chat through writing blocks and code blocks; paid users could keep canvas on legacy models only until those models were retired. For hosted pages and apps, OpenAI now has Sites, in public beta. Claude’s artifacts have moved the other way: they now include Docs, Slides and Design templates, storage, connected apps, and sharing with named people. In short, Claude keeps one object you make, keep and share; ChatGPT splits that job between drafts in the chat and hosted Sites. Neither is better in general. The sections below show which fits which job.

What happened to Canvas

Canvas opened a separate window beside the chat for writing and coding. You could highlight a passage to point ChatGPT at it, and use shortcuts for suggested edits, length, reading level and a final polish. OpenAI has not renamed it. It removed canvas from its current models and moved the parts people used into the conversation itself. A writing block is an editable area for a draft such as an email, a post or a document. A code block keeps code apart from the rest of the answer so you can edit, copy, preview or run it. What you can do with a block varies by plan, device, workspace settings, model and rollout, and the help centre does not list those differences in full.

So “Artifacts vs Canvas” is now really two questions. For drafting inside a conversation, compare artifacts with writing and code blocks. For something you publish and other people use, compare artifacts with Sites.

Editing and versions

• Claude. You can ask Claude to change an artifact, or, in one made from a template, type in the doc, edit a slide or move elements on a design canvas yourself. In a doc or Markdown document you highlight text and choose “Edit with Claude”. Editing an earlier message creates a new version of the chat with its own artifacts, and in Claude Code each publish becomes a version and you choose which one viewers see.
• ChatGPT. You edit a writing block directly in the chat, and the conversation carries on above and below it. A Site keeps saved versions: ChatGPT builds a version you can review, and only deploying it makes it live. A follow-up to an interactive visualisation can produce a replacement rather than an edit in place.

Running code

• Claude. An artifact can be a single-page website, a diagram, a dashboard or a small interactive tool, and it can call Claude, so it behaves like a small app for whoever opens it. A Claude Code artifact is one HTML or Markdown page, which may load libraries from a short list of public CDNs, with no backend.
• ChatGPT code blocks. OpenAI’s help centre lists previews for HTML pages, React components, SVG images, Mermaid diagrams and Vega or Vega-Lite charts, and a Run button for supported Python that opens a code workspace with a console. Previews and execution run in a sandbox.
• ChatGPT Sites. A full hosted application: according to OpenAI’s Sites documentation↗ (opens in a new tab), a Site can keep structured data in a database, store uploaded files, add Sign in with ChatGPT, use a custom domain you own, and show visitor analytics.

Sharing and publishing

Claude artifacts start private. Anthropic’s page on sharing artifacts↗ (opens in a new tab) lists “Anyone with the link” on paid plans, invitations by email in beta, and on Team and Enterprise plans everyone or named people in the organisation, where an owner decides whether anything can go outside it. Access is Can view, Commenter or Can edit, depending on the artifact. The help centre says everyone opening a shared artifact needs a Claude account, apart from legacy artifacts published from a chat, although Claude Code’s artifact pages describe links anyone can open without signing in; check the page for where you made yours. Artifacts that use Claude or your connected apps cannot be shared by link. How far these options go for a team is covered in Claude Artifacts sharing limits.

In ChatGPT, a writing or code block travels with its chat: you share the conversation. A Site has its own audience setting. It starts limited to its owner and workspace admins, and can be opened to selected people or groups, invited outside viewers, the whole workspace, or anyone on the internet where public publishing is enabled. In Enterprise workspaces an admin must turn public publishing on.

Collaboration

• Claude. The Docs template is a living document you write with Claude and your team in real time. People with Commenter or Can edit access can comment, and in Claude Code, Team and Enterprise users can send a comment thread to Claude and have it reply or make the change.
• ChatGPT. Site collaboration needs a workspace: the owner can make active members of that workspace editors, who can refine the Site and publish updates once the owner has published it the first time. Editors cannot change the audience, invite people or restore an earlier version, and they can read the Site’s live database.

Plans, named only

• Claude artifacts: Free, Pro, Max, Team and Enterprise. Templates are in beta on paid plans; storage and connected apps are on Pro, Max, Team and Enterprise. Code execution and file creation must be on.
• ChatGPT writing and code blocks: available actions depend on plan, device and workspace settings.
• ChatGPT Sites: public beta on Plus, Pro, Business, Enterprise and Edu, with plan-specific usage limits during the beta.

Claude artifacts vs ChatGPT Sites

This is the sharper comparison now. A Site is a deployed web application with a backend: routes, a database, file storage, sign-in, a custom domain and analytics. A Claude artifact is a page that lives inside Claude: it can remember text, call Claude and read the viewer’s own connected apps, but Anthropic’s Claude Code artifacts documentation↗ (opens in a new tab) is plain that it has no backend, cannot serve several routes and cannot authenticate viewers itself, and suggests deploying a hosted internal tool on your own infrastructure. If the thing must outlive the chat as a real website for people outside your tools, Sites is closer to the job. If it is a working page for people who already use Claude, an artifact is less to manage.

OpenAI also has Visualizations, a preview that is still rolling out: interactive charts, maps, diagrams and calculators inside a chat. Its Visualizations page↗ (opens in a new tab) describes each one as a snapshot rather than a live dashboard, and suggests turning one into a Site when it should be hosted and revisited.

What each is best at

• A document or deck to finish and send: Claude’s Docs and Slides templates, which export to Word, PDF, Google Docs and PowerPoint.
• A quick draft you will paste elsewhere, such as an email: a ChatGPT writing block, or a plain Claude reply. Neither needs a separate window any more.
• Running a bit of Python to check a result: a ChatGPT code block with Run, or Claude with code execution on.
• A small tool or dashboard for colleagues who use Claude, perhaps calling Claude or their own connected apps: a Claude artifact.
• A public site, internal app with sign-in, or anything that needs its own database and domain: a ChatGPT Site, or your own hosting.

Neither is a work list

Both products now make it easy to build a tracker, and both warn you to check who can see shared data before typing anything private into one. What neither gives a team is a list of work with members, roles and a record of who changed what. That is the part fenbs does: tasks that are features, enhancements or bugs, in lanes To Do, Next Up, In Progress and Completed, with Claude and ChatGPT connected over MCP as members acting under a person’s role. Build the dashboard in whichever tool fits; keep the list on the board, and let the dashboard read from it. A fenbs board is shared with its members only; it is not something you publish.