What happened to ChatGPT Canvas, and what an AI canvas can be

If you are searching for ChatGPT Canvas and finding a different product than you remember, you are not imagining it. On May 28, 2026, OpenAI updated GPT-5.5 Instant and GPT-5.5 Thinking and removed Canvas from both, replacing it with inline writing blocks and code blocks that live directly in the chat thread. Canvas survived only on legacy models, and those are leaving too: GPT-4.5 retired on June 27, 2026, and o3 follows on August 26. For practical purposes, ChatGPT Canvas is gone.

This post covers what Canvas was, what OpenAI replaced it with, and a question worth asking while you are choosing a replacement: what should a canvas for AI work actually be?

What ChatGPT Canvas was

Canvas opened a document or code file in a panel beside the chat. The conversation stayed on one side; the artifact you were building stayed pinned on the other, and edits landed in place instead of scrolling away. For iterating on one piece of writing or one file, it was genuinely useful, which is why its quiet removal frustrated the people who had built habits on it.

What replaced it

Inline writing blocks and code blocks: the same editable artifacts, but embedded in the flow of the conversation rather than pinned beside it. If Canvas was how you edited one document with the model, the blocks cover most of that. OpenAI folded the feature back into the thread rather than maintaining a second surface.

What the blocks do not change is the shape of the conversation around them. ChatGPT remains a single scrolling thread: one line of conversation, with everything you explored earlier disappearing upward, and the context the model sees assembled invisibly from a hidden system prompt, synthesized memory, and however much of your history still fits.

The word “canvas” was always bigger than the feature

Here is the thing the feature name gestured at but never delivered: a canvas is supposed to be a surface where you can see your work laid out and move around it. ChatGPT Canvas pinned one artifact next to one linear chat. The conversation itself, the part where your actual thinking lives, stayed a scroll.

A real AI canvas treats the conversation as the thing on the canvas. Every message is a node you can see. When a side question threatens to derail a thread, you fork from that exact point and both paths stay visible, side by side, each carrying the full history behind it. When two threads each hold a piece of the answer, you merge them into scope and draft from both. When the model answers, it answers from exactly the context you assembled, and nothing else.

That is what we build NodePad to be, so discount our enthusiasm accordingly. But the distinction is real regardless of whose product you pick: a document beside a chat is an artifact viewer; a conversation you can see, branch, and merge is a canvas.

What to use now

• You mainly edited one document or file at a time. Stay in ChatGPT; the inline writing and code blocks are the direct successor, and they are on the current models.
• You want the pinned-artifact workflow elsewhere. Claude’s Artifacts and Gemini’s Canvas both keep a live artifact beside the chat, and both are active features, not sunset ones.
• What you actually wanted was room to think in more than one direction. That is the job the word promised, and it is what NodePad is for: your chats on one infinite canvas, forkable at any point, mergeable when threads converge, with Claude, GPT, and Gemini picked per message. There is a live demo on the homepage you can try without signing up.

We keep a fuller migration guide at NodePad vs ChatGPT Canvas, including a feature-by-feature table, and an honest roundup of eight Canvas alternatives if you want the whole field.

One quiet lesson from the episode is worth keeping regardless of what you choose: features inside a single vendor’s chat app can disappear on that vendor’s schedule. Surfaces you build real habits on deserve to be products, not features.