You ask ChatGPT to split a dinner bill with friends. It could respond with a formula and a short explanation. But what if the answer already included controls to change the group size, include a service charge, and see everyone's share?
That is the sort of interaction OpenAI wants to make commonplace with Intelligent UI, announced on October 7 as GPT-6 expanded across ChatGPT plans, including Free.
The point is not to make text more colorful. It is to turn an answer into something you can actually use.
Ask a question. Get controls when they help
According to OpenAI's announcement, GPT-6 can decide how to arrange a response. Some questions need a paragraph. Others benefit from side-by-side comparisons, maps, charts, forms, or small interactive tools.
The company's examples illustrate the idea:
- Everyday finances: a calculator for exploring how savings might grow.
- Learning: a visual explanation that changes as you adjust a problem's inputs.
- Travel planning: a map-based itinerary with stops and suggestions.
- Quick decisions: a bill splitter that does not require opening a spreadsheet.
You do not need to know programming or write an elaborate prompt. Something as simple as "Help me split this bill among five people" can be enough.
TRY THE DIFFERENCE
The same question, two experiences
An illustrative demo made by MnzAI Labs. The values and controls below are not an official ChatGPT screenshot.
R$198.00
Subtotal: R$180.00. Editorial example with locally computed, fictional amounts. It is not a live GPT-6-generated interface.
How ChatGPT builds an interface while answering
The technology is more interesting than the visual polish. OpenAI describes a library of native components and a compiler that processes an interface as the model generates it. Text, buttons, and visuals can appear progressively instead of waiting for the entire response.
The model was also trained to decide when each component is useful. That matters: nobody needs a six-control dashboard just to learn a country's capital.
GPT-6 can also start responding while it continues reasoning. In an internal OpenAI evaluation, Instant began answers involving web search 44% earlier on average than GPT-5.6 Instant. This is a vendor-reported result, not an independent measurement of speed for every user.
Didn't Gemini and Claude already do something similar?
Yes, and that context matters. This competition did not start in October.
| Product | Where the experience appears | Main approach |
|---|---|---|
| ChatGPT with Intelligent UI | Interactive components inside the answer | Native, progressive interface selected for the question |
| Gemini with Dynamic View | An interactive view designed and coded by the model | Custom UI generated through coding capabilities |
| Claude with Artifacts | Tools, documents, and apps in a dedicated area | Standalone, editable, and often shareable content |
Google had already introduced Dynamic View in Gemini. Anthropic offers Claude Artifacts, including interactive apps.
OpenAI's stated distinction is how the interface fits into the normal conversation flow, with native components and progressive rendering. That alone does not establish that any one product is more accurate or more versatile.
Who can use it?
Rollout began on October 7, 2026 for Plus, Pro, Business, and Enterprise, and expanded to Free and Go on October 8. Because rollout is gradual, accounts on the same plan may not see identical features immediately.
In ChatGPT, the paid plans listed use GPT-6 Sol for everyday chat, while Free and Go use GPT-6 Luna. This announcement applies to Chat, and OpenAI explicitly said the models powering Work and Codex are not changing with this release.
That distinction matters if you followed our coverage of the GPT-6 Sol and Luna model family. This story is about interaction, not the API pricing table.
A polished interface can still be wrong
AI-generated interfaces bring familiar risks in a new package. A persuasive chart may use incorrect numbers. A calculator may misinterpret a rule. A form may omit an important condition.
Check data sources, assumptions, and calculations, particularly for financial, medical, or legal decisions. Privacy controls and permissions still matter in the workplace. A pleasant interface is not a substitute for verification.
Nor should a tool created in the chat be confused with production-ready software. Systems integration, persistent storage, testing, accessibility, and operational responsibilities still require dedicated work.
When software begins adapting to people
For decades, we learned our way around software designed in advance, figuring out which menu to open and which button to click. Generative interfaces partly reverse that arrangement.
Instead of hunting for a tool, you describe your problem. The AI attempts to assemble a useful experience on the spot.
The important change, in MnzAI Labs' view, is not replacing paragraphs with cards. It is shortening the distance between understanding an answer and acting on it. If this becomes reliable, accessible, and trustworthy, the conversation will be more than the place where we ask for help. It may also become the place where we solve the problem.