Canvas (ChatGPT)
Last reviewed
Sources
6 citations
Review status
Source-backed
Revision
v1 · 1,567 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
6 citations
Review status
Source-backed
Revision
v1 · 1,567 words
Add missing citations, update stale details, or suggest a clearer explanation.
Canvas is an interface for ChatGPT, the conversational artificial intelligence product from OpenAI, that opens a writing or coding project in a separate window alongside the chat thread. Rather than working only through the back and forth of a conversation, users can edit text or code directly in the canvas window, highlight specific passages for targeted changes, and iterate on a single document over time. OpenAI introduced canvas in beta on October 3, 2024, and made it generally available to all ChatGPT users, including those on the free tier, on December 10, 2024, during the company's "12 Days of OpenAI" series of announcements.[1][2][3]
Canvas is a user interface feature rather than a model. The behaviors that drive it, such as deciding when to open a canvas or how to apply an edit, were trained into the underlying GPT-4o model; the canvas window itself is the presentation layer through which that collaboration takes place.[1]
Canvas addresses a limitation of the standard chat format, in which a long document or code file produced by ChatGPT can only be revised by sending further messages and receiving a new copy of the whole output each time. When canvas is active, ChatGPT's response appears in a panel beside the conversation, and the text or code becomes a persistent, editable object. Users can type into it directly, select a portion of the content to comment on or change, and ask ChatGPT to rewrite only the highlighted section while leaving the rest untouched.[1][4]
OpenAI described canvas as the first major change to ChatGPT's visual interface since the product launched in 2022, and as "a new way of working together, not just through conversation, but by creating and refining content side by side."[1] Daniel Levine, a product manager at OpenAI, characterized the side window as "a more natural interface for collaborating with ChatGPT."[2] The feature is broadly comparable to Anthropic's Artifacts, released in June 2024, and to Google's Canvas in Gemini, both of which present generated content in a dedicated workspace.[2]
Canvas includes a version history. Toolbar arrows let a user move between earlier and later versions of the document, a "show changes" control (represented by a clock icon) marks deletions in red and additions in green, and a "restore this version" option reverts the canvas to a previous state.[4] These controls give the side window some of the affordances of a conventional editor, so that revisions can be reviewed and undone rather than lost in the flow of a conversation. A finished canvas can also be copied out of the window for use elsewhere.[4]
Canvas can be opened in two ways. ChatGPT may open one automatically when it judges that a request, such as drafting a long document or working through a coding problem, calls for a dedicated workspace. A user can also trigger it explicitly, for example by typing a phrase such as "use canvas." During the beta period a "GPT-4o with canvas" entry in the model picker served the same purpose, an extra step that was removed once the feature became part of the default GPT-4o experience.[1][2][3]
For prose, canvas exposes a menu of one click shortcuts that appears at the lower right of the window, in addition to free form requests typed into the chat. ChatGPT can also make inline suggestions that the user accepts or rejects, and the length of a document can be adjusted along a slider.[1][4] The writing shortcuts are summarized below.
| Writing shortcut | Function |
|---|---|
| Suggest edits | ChatGPT adds inline suggestions and comments to improve the text |
| Adjust the length | Shortens or lengthens the document |
| Change reading level | Adjusts the reading level along a range from Kindergarten to Graduate School |
| Add final polish | Checks grammar, clarity, and consistency for a finished draft |
| Add emojis | Adds emojis for emphasis and color |
Reviewers have generally welcomed the document style workspace for tasks such as drafting and revising, noting that it removes the friction of repeatedly copying text in and out of the chat. Critics at launch pointed to a limited editor and the absence of real time co-editing between multiple people.[5]
Canvas applies the same side window model to programming. Code is displayed with syntax highlighting, and a separate set of shortcuts supports common editing and debugging actions. As with writing, a user can highlight a span of code and ask for a localized change.[1][4] The coding shortcuts are summarized below.
| Coding shortcut | Function |
|---|---|
| Review code | Provides inline suggestions to improve the code, which the user can approve |
| Add logs | Inserts logging or print statements to aid debugging |
| Add comments | Adds explanatory comments to the code |
| Fix bugs | Detects and rewrites problematic code to resolve errors |
| Port to a language | Translates the code into JavaScript, TypeScript, Python, Java, C++, or PHP |
With the December 2024 general availability update, canvas gained the ability to run Python code directly inside the canvas window and display the results, including text output and generated graphics, without leaving ChatGPT. ChatGPT can read console errors from a run and propose fixes, creating a tighter edit and test loop. The execution happens in the browser through a WebAssembly based Python environment, which is distinct from the server side Code Interpreter tool that ChatGPT also offers.[2][3][6]
At launch on October 3, 2024, canvas was a beta feature reserved for paying tiers, and users selected "GPT-4o with canvas" from the model picker or typed a request such as "use canvas" to invoke it; ChatGPT could also open a canvas automatically when it judged that a task, such as a long document or a coding problem, would benefit from one. With the December update the feature was folded into the default GPT-4o experience, removing the separate model picker entry, and was extended to every ChatGPT account. At that point canvas was available on the web and on the ChatGPT desktop app for Windows.[1][2][3] The principal milestones are summarized below.
| Date | Milestone |
|---|---|
| October 3, 2024 | Canvas launches in beta for ChatGPT Plus and Team users worldwide, built with GPT-4o[1] |
| Week of October 7, 2024 | Access extended to ChatGPT Enterprise and Edu users[1][2] |
| December 10, 2024 | General availability on Day 4 of "12 Days of OpenAI"; rolled out to all users including the free tier, enabled by default in GPT-4o, with in browser Python execution and support inside custom GPTs[3][6] |
The December release also brought canvas to custom GPTs, the user built versions of ChatGPT. Canvas is enabled by default for newly created custom GPTs, while existing ones can have it turned on through their configuration.[3][6]
Canvas is a product surface built on top of GPT-4o rather than a separate model. OpenAI has said it trained GPT-4o to act as a collaborator that knows "when to open a canvas, make targeted edits, and fully rewrite," and to decide when a request is better served by an inline chat reply than by opening the side window.[1] When a canvas is open, the model can operate on the document in several ways: rewriting the entire contents, applying a change to only a highlighted selection, or attaching inline suggestions and comments that the user reviews. The one click shortcuts map to predefined instructions for these operations, such as shifting the reading level or porting code to another language.[1][4]
This design distinguishes canvas, the editable workspace and its controls, from the model that performs the work. The same GPT-4o model powers ordinary ChatGPT conversations; canvas adds a structured, persistent document and a set of editing affordances around it. The in browser Python feature added in December is likewise a capability of the canvas surface, separate from the cloud based Code Interpreter sandbox.[1][2][6]
Coverage of canvas was largely positive, framing it as a meaningful step in making ChatGPT useful for sustained writing and coding work rather than one off answers. Technology press noted that the side by side layout reduced the copy and paste friction of the chat only workflow and drew comparisons to Anthropic's Artifacts and to coding focused tools.[2][5] Hands on reviewers praised the convenience of the shortcuts and version history for drafting and editing, while some described the early editor as limited or occasionally glitchy and pointed to the lack of multi user co-editing as a gap relative to collaborative document tools.[5] Observers also placed canvas within a broader trend, accelerating through 2024, of AI chat assistants adding dedicated workspaces for generated artifacts.[2][3]