OpenAI Unveils 'Canvas' for ChatGPT: An Enhanced Interface Designed for Writing and Coding Projects

OpenAI unveiled a new and innovative way to engage with ChatGPT on Thursday: the “canvas” interface. This feature introduces a separate workspace adjacent to the typical chat window, designed for writing and coding projects. Users can generate writing or code directly within the canvas, easily highlighting sections for the model to edit. The canvas is launching in beta for ChatGPT Plus and Teams users today, with availability for Enterprise and Edu users next week.

As consumer AI providers increasingly adopt editable workspaces, this trend provides a practical approach to utilizing generative AI effectively. ChatGPT's new interface parallels offerings like Anthropic’s Artifacts, which debuted in June, and the popular coding assistant, Cursor. OpenAI is actively enhancing ChatGPT to compete with rival products while also introducing new features aimed at expanding its paid user base.

Currently, AI chatbots struggle to complete extensive projects based on a single prompt, but they are able to deliver solid starting points. Workspaces like canvas enable users to amend any inaccuracies in the chatbot's output effortlessly, without the need to revise their original prompts or generate entirely new code segments.

"This interface creates a more intuitive experience for collaborating with ChatGPT," explained OpenAI product manager Daniel Levine during a demonstration.

During our demo, Levine chose “GPT-4o with canvas” from ChatGPT's model selection dropdown. However, OpenAI mentions that canvas windows will automatically appear when ChatGPT identifies that a separate workspace might be beneficial—this is particularly useful for lengthy outputs or intricate coding tasks. Alternatively, simply writing “use canvas” will trigger the project window to open.

Levine demonstrated how ChatGPT's new features can assist in writing an email. Users can request ChatGPT to draft an email, which will promptly appear in the canvas window. They can then adjust the length of the text using a slider, or highlight specific sentences to request modifications like “make this sound friendlier,” or enhance the message with emojis. Additionally, users can instruct ChatGPT to rewrite the entire email in another language while preserving the original intent.

The coding canvas offers distinct features. Levine prompted ChatGPT to build an API web server in Python, which generated the code within the canvas. By clicking an “add comments” button, ChatGPT can incorporate in-line documentation to clarify the code in plain language. Users have the ability to highlight a segment of code that ChatGPT has produced to request explanations or pose questions about it. A new “review code” button will suggest specific edits for any code present in the window—whether user-created or AI-generated—allowing users to approve, modify, or reject the suggestions. If approved, ChatGPT will attempt to correct any identified bugs.

Once the canvas feature exits beta, OpenAI has announced plans to make it accessible to free users as well.