ChatGPT canvas has been missing from the models you use by default since 28 May 2026. OpenAI said in its release notes that canvas is no longer available in GPT-5.5 Instant and GPT-5.5 Thinking, and that writing and coding now happen in the chat itself through writing blocks and code blocks.
If you knew canvas as the separate workspace to the right of the conversation, you will mainly notice what is no longer there. There is no separate panel and no shortcuts menu for writing length or reading level. Instead of the back button with version history, you can undo or redo recent AI edits in a writing block, and you can open a block full screen if you want. The functionality has not disappeared completely, but it has become flatter and moved into the thread, just as happened earlier with other parts of the interface, including ChatGPT agent mode.
For paid accounts, canvas remained accessible for a while through older models, but OpenAI is switching those off one by one. If you had built a writing process around it, you are better off setting it up differently now.
Does ChatGPT canvas still exist?
ChatGPT canvas no longer exists in the models most people use. On 28 May 2026, OpenAI removed the feature from GPT-5.5 Instant and GPT-5.5 Thinking, the default models for free and paid accounts at the time. Those models have since been succeeded, and writing blocks have remained the standard in each of their successors. Since then, writing and coding have run through writing blocks and code blocks in the chat.
The distinction that remains is the one between current and legacy models. Paid users can still select older models in the model picker, and canvas does work in those, but OpenAI explicitly calls this a temporary situation until those models are switched off. Free accounts do not have that option.
The confusion about this is understandable. Canvas launched in October 2024 as a beta for Plus and Team users, then became more widely available and disappeared again without a separate announcement, mentioned only as a line in the release notes for a model update. Many guides online still describe the situation before May 2026.
If you are not sure whether you still have canvas, one quick test will tell you. Ask for a longer text and watch where the answer appears. If a panel opens to the right of the conversation, you are on a legacy model. If you get an editable block in the thread itself, you are working with the new set-up.
What are writing blocks and code blocks?
Writing blocks and code blocks are the successors to canvas and work in the chat itself rather than in a separate panel. You get an editable block in the thread where the answer would normally appear, and you edit that block without the conversation pasting a new version underneath.
The reason OpenAI gives is consistency across devices. A side panel works differently on a wide screen than on a phone, whereas a block in the thread is the same everywhere. That also explains why the feature has become flatter: everything that depended on the space next to the conversation has gone with it.
In practice, it means you steer edits through the chat again. You ask for a change, the block is updated and the conversation carries on around it. What remains is that the model does not churn out your whole text again with every change, which was the main gain of canvas.
In terms of formatting, little changes. Writing blocks support bold, italics, headings, links, bullet points, numbered lists and checklists. Canvas did not offer more extensive formatting either, so if you ran into its limits then, you still will. With code, ChatGPT detects the programming language automatically as soon as you paste or generate a snippet. Code blocks show a preview for HTML, React, SVG and Mermaid, among others, and you can run supported Python code in a code workspace.
What carries over between conversations falls under a different feature. Instructions and preferences you use often belong in ChatGPT memory, which is separate from whether you work in a block or a panel.
What could canvas do that writing blocks cannot?
Canvas offered three things that are missing or weakened in the current set-up. It pays to know which, because then you also know what you now have to handle yourself.
It started with how it opened. Canvas opened automatically as soon as ChatGPT detected that you were starting a writing or coding project, and you could also call it up yourself by asking for it or clicking the canvas icon. That step no longer exists, because a writing block just appears in the answer.
Version control and targeted edits
In canvas, version control sat in a back button in the toolbar, which let you restore an earlier version of your document or code if an edit went the wrong way. That history lived in the canvas itself, not in the chat, so you did not have to scroll back to see what had changed.
In canvas you could highlight a specific piece of text and have only that part edited, or ask a question about it. You can still do that in a writing block: you select text and have ChatGPT rewrite only that part, or you edit the text directly in the block. Do save important interim versions yourself, outside ChatGPT, because undo only works for recent AI edits.
The shortcuts menu
The shortcuts menu sat at the bottom right of canvas and contained fixed edits: suggest edits, adjust the writing length, change the reading level or add a final polish. For code there was a version with code review, bug fixing and showing suggestions.
The buttons are gone, the underlying tasks are not. You can ask for everything the menu did in a prompt, only now you have to write that prompt yourself. When I used ChatGPT intensively, that menu was mainly a time-saver, not a capability you could not get any other way.
Exporting and sharing
In canvas you exported via a download button at the top right, which sent a document out as a PDF, Markdown or Word file. With code, canvas recognised the programming language and automatically chose the right extension, such as .py, .js or .sql.
Sharing worked much like sharing a conversation: you sent on a canvas item, and the recipient saw it as you had left it. You now send an email draft from ChatGPT where available, or open it in your email app, and you save a document draft in your Library. For anything else, you copy the content from the block and put it into a file or document yourself, which adds a step for work you share often.
If you regularly deliver documents, the difference in daily practice lies mainly in the download button. OpenAI does not mention export as PDF, Markdown or Word for writing blocks. If the text goes out as an email, send it from the block or open it in your email app. Keep a draft you will develop later in your Library. If the text has to live in a document or system of your own, copy the content of the block into the program where it will end up anyway. With code, you paste the snippet straight into your editor, which recognises the language just as well as canvas did.
Can you still use canvas through legacy models?
Canvas still works in older models, but only for paid accounts and not for much longer. OpenAI removed GPT-4.5 from ChatGPT on 26 June 2026 and OpenAI o3 on 26 August 2026. That means the best-known places where canvas still ran have already gone.
In practical terms, that means one thing: do not build a way of working on it any more. A process that depends on a feature in a model that can disappear from the model picker at any moment will have to be converted anyway. Better to do it now, while you have time to test it properly.
Which models you can choose at all depends on your subscription, and the free version has no model picker with legacy options.
You can see which model you are currently using in the model picker above the conversation. If it says GPT-6, you are working with writing blocks. Under the bonnet, GPT-6 Sol runs on paid plans and GPT-6 Luna on Free and Go. If you select a legacy model, you get canvas back, but every new conversation you start without that selection begins on the current model again. That makes it hard to work with canvas consistently.
What should you use instead of ChatGPT canvas for longer documents?
Instead of canvas, you work on longer documents with writing blocks, combined with a place where you keep context permanently. Canvas worked per document, and that was also its weakness: every new text started without the instructions and files from last time.
For recurring work, a ChatGPT project is the more logical place. Files, instructions and conversations stay linked to one topic there, so you do not have to explain the context again in every session. For ways of working you want to hand to others, you could create a custom GPT, but OpenAI is phasing that feature out from 11 December 2026 and now only allows new ones to be created within a Business, Enterprise or Edu workspace. To me, custom GPTs felt more like a watered-down version of Claude projects, with less consistent output between testing and daily use.
Outside ChatGPT, the closest alternative is Claude artifacts, which still work as a separate panel next to the conversation, where you keep documents and code in a window of their own. When I used ChatGPT intensively, I noticed with long documents that Claude lost less context than ChatGPT, especially with texts I refined over several sessions.
I briefly tested ChatGPT 5.5 after its release to see where the model stands among other LLMs. The step from 5.4 to 5.5 felt more like refinement than a breakthrough, and the disappearance of canvas fits that picture: less an improvement in what the model can do, more a simplification of the interface around it.
The choice between those three depends on how often the work comes back. Write one-off texts in a writing block, with no further set-up. Work that returns every week with the same starting points belongs in a project. Work a colleague should also be able to do without your explanation belongs in a Plugin or in a documented prompt the team shares.
What does this mean for the way you work?
The disappearance of canvas shows how fragile a way of working is when it leans on one feature from one vendor. If you had organised your writing process around the canvas panel, the back button and the shortcuts menu, you had to rethink that set-up after a single line in the release notes of a model update. Some of it came back in another form. You undo or redo recent AI edits in a writing block, and you can have a selected piece of text rewritten separately. A full version history and the shortcuts menu did not come back.
The lesson lies in the distinction between task and tool. What you need is a document you refine step by step, versions you can restore and a check before the text goes out. The button that gives you that changes from quarter to quarter. Record the steps, not the buttons, and keep interim versions in a place you manage yourself.
In practice, that comes down to three habits. Keep interim versions of important texts outside the tool, in a folder or document environment you manage yourself. Write down the steps of your process somewhere, so you know which actions you need to find again if a button disappears. And check the output before it goes out, whatever panel or block it was created in.
An approach like this also makes it easier to switch when a vendor takes something away. The step stays the same, only the tool changes, and then the transition is just a matter of finding out where the button is now.
If you want to learn to use ChatGPT more broadly than one feature, with prompt structure and output checking as a fixed way of working, you will find that structure in LearnLLM’s ChatGPT Course.
Frequently asked questions about ChatGPT canvas
Now that ChatGPT canvas has gone, questions remain about alternatives outside ChatGPT and about what code blocks can do.
Does Gemini have a canvas too?
Yes, Gemini has its own Canvas, a workspace for writing and programming that Google says is available to all Gemini users, including on the free plan. You open Canvas by selecting it below the input bar and then typing your request. In the mobile app you can open Canvas projects, but you can only edit text style and formatting in the web version of Gemini on a computer.
Can you run code in a ChatGPT code block?
Yes, you can run supported Python code from a ChatGPT code block in a code workspace, which shows the output or error messages. For HTML, React, SVG, Mermaid diagrams and Vega charts, you switch between a code view and a preview. According to OpenAI, previews and execution run in an isolated environment, and a workspace admin can switch off execution and network access.
