On October 7, 2026, OpenAI announced that it is gradually rolling out GPT-6 and a new feature called "Intelligent UI" to the Chat tab in ChatGPT. Depending on the question, answers combine text and diagrams, and can also include small tools with buttons and input fields. ChatGPT has long had charts and editing screens, but this time the model itself chooses how to present the answer to an everyday question. If you see this only as one more way to display diagrams, it is easy to miss how it differs from existing features and what conditions apply to using it.
Diagrams and small tools tailored to the question
![FireShot Capture 001 - GPT-6 and Intelligent UI for everyone - OpenAI - [openai.com].webp](https://media.xenospectrum.com/Fire_Shot_Capture_001_GPT_6_and_Intelligent_UI_for_everyone_Open_AI_openai_com_dfbbfdd0f2.webp)
With Intelligent UI, ChatGPT structures an answer to fit the question: items are laid out side by side for a comparison, and an interactive diagram is used to explain how something works. OpenAI's examples include a cooking plan that combines steps and timing, and a map showing stops on a trip. The company also says ChatGPT can build small tools for use within a conversation, such as a savings calculator or a tool for splitting a restaurant bill with friends.
Reading an explanation and experimenting with different conditions are not the same thing. If you only receive a calculated result, you have to rewrite your question to change the assumptions. With a calculator you can type into, you can compare conditions and results on the same screen. Building this kind of interaction into answers is what Intelligent UI aims to do. However, the published examples do not demonstrate improved learning efficiency or fewer wrong answers.
Text answers remain. According to the official help page, no special prompt is needed, and visual elements and interactive components are combined when they are useful. The design does not turn everything into diagrams or forms, even for short questions that deserve a concise reply.
How does it differ from Canvas and data analysis?
Canvas, which OpenAI introduced in October 2024, is a feature for collaboratively editing text and code in a separate window. You can select the part you need, ask for revisions, and rewrite it directly yourself. Data analysis can create charts from uploaded tables and display interactive tables and charts. This is not the first time ChatGPT can handle more than text.
What is new about Intelligent UI is less that it can display diagrams and more that the model chooses the structure of the answer and how you interact with it, based on the question.
Classifying OpenAI's official descriptions by purpose and by who builds the screen shows the following differences.
| Feature / approach | Main purpose | How the screen and interaction are built |
|---|---|---|
| Canvas | Collaboratively editing text and code | A separate editing screen, with selection of parts, direct editing and revision suggestions |
| Data analysis | Calculating and exploring data with tables and charts | Analyzes uploaded data and uses Python for calculations as needed; interactive tables and charts are also available |
| UI in plugins | Comparing, editing and checking information from external tools | A UI prepared by the developer is returned by an MCP server and shown within the conversation |
| Intelligent UI | Answering everyday questions in a suitable form | The model combines text and screen components, composing diagrams and small tools as the answer |
This is a classification of roles based on public descriptions as checked on October 8, 2026, not a table comparing speed or accuracy. Canvas is taken as it was designed at launch. Because Canvas could already open automatically when useful, the model deciding whether to show a screen cannot be called a first for this release.
The difference lies in why the screen is created. For revising a manuscript, Canvas handles the object being edited; for examining a sales table on hand, data analysis fills that role. Intelligent UI builds the explanation and interaction needed for the question at hand into the answer. This was not announced as a replacement for existing features, so there is still room to use each depending on what the user wants to investigate and operate.
Developer-built UI and model-composed UI should also be told apart. In plugins, developers prepare UI resources and connect them to processing in external tools. The description of Intelligent UI, by contrast, centers on the model composing the answer. The announcement does not say whether the two share the same component library or execution method.
The 44% speedup applies to when answers start
For Intelligent UI, OpenAI built a library of screen components inside ChatGPT and a compiler that processes screens while they are being generated. Because display can begin while the model is still assembling components, there is no need to wait until the whole screen has been generated. The company says it also trained the model on how to lay out text and diagrams and when to use interactions.
The model is responsible here not only for content but for the judgment of presentation. Even with the same information, a useful layout differs between a step-by-step explanation and a comparison of items. Shared components support the foundation of the screen, and the model decides how to combine them. However, details have not been disclosed, such as where calculations happen each time a user changes conditions, or whether the model reasons again.
GPT-6 can also keep thinking while giving an initial answer, then add more detailed content afterward. According to OpenAI, for questions that require web search, GPT-6 Instant starts answering on average 44% faster than GPT-5.6 Instant. This is the company's evaluation of the time until an answer begins, comparing Instant models on questions that need search.
The 44% figure does not indicate the time until the whole answer is complete. The announcement also gives neither absolute times for the comparison nor the sample size, so it cannot be said that every question will see the same reduction. A screen or text appearing sooner changes how waiting feels, but the correctness of the final answer and the benefit of using diagrams need to be checked separately.
Distinguishing the Pro plan from the Pro thinking setting
In the October 7 announcement, GPT-6 and Intelligent UI go first to Plus and Pro, with Business and Enterprise also eligible from the same day. The plan is to extend them to Free and Go from the 8th, with paid plans using GPT-6 Sol and Free and Go using GPT-6 Luna. All of this is a staged rollout, and on Enterprise, administrator settings also have an effect.
However, completion of the switch for free users cannot be confirmed. The official help page still contains a note that Free and Go use GPT-5.6 Luna. The rollout plan in the announcement and the model each account can currently select need to be distinguished. If GPT-6 does not appear in the model menu, it may not yet have reached that account.
Even on paid plans, behavior differs depending on the thinking setting you choose. According to the official help page, Intelligent UI and the feature that answers while thinking are available from Instant through Extra High, but not with the Pro thinking setting that uses Astra. Subscribing to the Pro pricing plan does not mean you can use the new UI if you set thinking to Pro.
There are conditions on the device side too. It works on the web and on supported, updated apps, and on mobile, updating to the latest version is advised. Older macOS and Windows versions of the ChatGPT app do not support this feature, and OpenAI recommends using the web version. This update targets the Chat experience, and the models for Work and Codex are not changed in this release.
Intelligent UI has no separate usage allowance; the limits of existing models, tools and plans apply. A change in appearance is not the same as an increase in usable volume. If you want fewer visual elements, you can turn off "Layout and visuals" in the web settings. Some visual elements may remain, though, so it is not a setting that returns to a fully text-only display.
Gemini is moving too: a race to build the answer screen
On May 19, 2026, Google announced a new design for Gemini called "Neural Expressive." It says answers are composed on the spot to fit the question, weaving in images, interactive timelines and dynamic diagrams. The direction of AI choosing how to present answers is not unique to OpenAI.
The two companies' announcements alone cannot settle which is better in the range of interactions or generation speed. Even so, a new question is added to comparing models. Beyond whether information is stated correctly, they are asked whether they can lay out items for comparison, show the relationship between changed conditions and results, and recognize situations where no extra interaction is needed.
A clear diagram does not guarantee that its premises are correct. For existing data analysis, OpenAI also advises checking the generated code, outputs and calculation assumptions. When using a calculator in Intelligent UI, too, you should confirm what you enter and under which conditions the result changes. A button that works and an explanation or calculation that can be trusted are separate matters to evaluate.
If users can change conditions and trace why results change, the answer screen can become a place to test one's own hypotheses. The measure of ChatGPT's new UI is less how many diagrams it shows than whether it provides the interactions and explanations that let you verify what you want to know for yourself.