GPT-6 Intelligent UI: generative UI goes mainstream in ChatGPT

Generative UI is now part of the most widely used AI product. With GPT-6, ChatGPT can answer with charts, tappable buttons and forms instead of only paragraphs, and the model picks the format that fits the question ^[1]. For designers, the interesting part is not the model. It is the component library underneath, because that is where the quality of every generated interface is decided.

OpenAI started rolling out GPT-6 with Intelligent UI on October 7 for Plus, Pro, Business and Enterprise users, followed by Free and Go users from October 8 ^[1]. Here is what shipped, how it compares with Google’s approach, and what we think design and product teams should do about it.

Key takeaways

• GPT-6 composes answers from text, visuals and interactive elements, rendered from a library of native components ^[1].
• Paid ChatGPT plans run GPT-6 Sol and Free and Go plans run GPT-6 Luna ^[1].
• Users can turn the amount of visuals up or down to suit their preference ^[2].
• Google takes a different route: its Gemini experiments have the model write the interface code itself ^[3].
• For product teams, a documented design system is the prerequisite for doing generative UI well.

What OpenAI shipped with GPT-6

According to OpenAI’s announcement, GPT-6 decides how to answer based on the question. It can build interactive visuals that show how something works, create small tools such as games, calculators or bill splitters inside the conversation, and explain concepts with inputs the user can adjust ^[1]. Answers also come with built-in controls for customising them ^[1].

“A comparison might work best side by side; an explanation might be easier to understand through an interactive diagram.”

TechCrunch’s Lucas Ropek describes tappable buttons, custom calculators, interactive charts and editable graphs that people can change directly in the chat ^[2]. Demo examples included a diagram of how an airplane wing generates lift, visual recipe guides, a multi-day hiking map and a personal savings calculator ^[2]. OpenAI product manager Aarush Selvan told TechCrunch that ChatGPT has mostly been a text interface, and that for many tasks the most helpful answers are not just text ^[2].

The rollout uses two models. Plus, Pro, Business and Enterprise get GPT-6 Sol, while Free and Go get GPT-6 Luna ^[1]. OpenAI also says GPT-6 Instant starts answering questions that need a web search 44% sooner on average than GPT-5.6 Instant ^[1]. That figure measures time to the first part of the answer, not the time to a complete one ^[1].

How generative UI works in ChatGPT

The technical detail that matters most for designers is short. OpenAI says GPT-6 can begin answering while it continues to think, and that a library of native components and a compiler display the interface as the model generates it ^[1]. In other words, the model does not invent new widgets. It chooses from a fixed set of parts and arranges them, and the interface appears progressively instead of all at once.

That is a deliberate trade. A closed component set keeps answers visually consistent, accessible in predictable ways and fast to render. It also limits how unusual an answer can look. Users who find the visual answers too busy are not stuck with them either, since they can reduce how much visual content appears ^[2].

Two approaches: component library or generated code

Google has been running a different kind of generative UI. In November 2025, Google Research described dynamic view, an experiment in the Gemini app where Gemini designs and codes a fully customised interactive response for each prompt using its agentic coding abilities ^[3]. The model generates HTML, CSS and JavaScript rather than filling a predefined template ^[3]. Google rolled this out with visual layout in the Gemini app and in AI Mode in Search for Google AI Pro and Ultra subscribers in the US ^[3].

Google’s researchers report that people preferred its generative UI output over other AI output formats by a wide margin, with only expert-designed websites rated higher, although that evaluation did not consider generation speed ^[3]. They are also candid about the costs.

“Our current implementation can sometimes take a minute or more to generate results, and there are occasional inaccuracies in the outputs.”

Put side by side, the two products show the real design question. Generated code gives the most freedom and the least predictability. A component library gives speed and consistency, at the price of range. Most client products we work on need the second option: an assistant inside a banking or booking app has to look like that app, every time.

Why the design system becomes the product

Nielsen Norman Group defined generative UI back in 2024 as an interface that AI generates in real time to fit the user’s needs and context ^[4]. Kate Moran and Sarah Gibbons argued then that designers would move from designing screens to designing outcomes, and warned that interfaces which change constantly can hurt usability because people lose the benefit of learned patterns ^[4].

“Humans will need to provide guidance and constraints for generative UI.”

GPT-6 is that guidance in practice. The model decides the composition, but the ceiling on quality is set by the components and by the rules about when each one fits. A chart with a poor empty state, a form without clear errors or a button with vague copy will be repeated in every answer that uses it. The design system stops being an internal tool and becomes the visible product.

Clients will soon ask for the same pattern in their own apps. A travel assistant that answers a date question with an availability picker. A finance assistant that replies with a small chart and a button to set a budget. Teams that already have well-documented components will be able to say yes quickly.

How your team should act now

Start with an audit of the components that an assistant would most likely use: charts, comparison tables, forms, pickers and confirmation dialogs. Check that each one works at small widths, streams in gracefully and still makes sense when it appears without the surrounding page.

• Write a one-paragraph purpose and a list of do and do not cases for every component.
• Design partial and progressive states, since generative interfaces render while the answer is still forming ^[1].
• Keep a plain text fallback and let users choose it, as ChatGPT does ^[2].
• Test generated compositions with real users, because consistency and learnability are the known risks ^[4].

Finally, decide early where your product sits on the spectrum between a fixed component set and generated code. For brand-sensitive and regulated products, we recommend the component approach. For learning tools and prototypes, generated code can be worth the slower speed.

FAQ: generative UI in ChatGPT

What is generative UI? It is an interface that an AI system creates in real time to fit the user and the task, rather than a fixed screen designed in advance ^[4]. In ChatGPT it means answers that mix text with charts, buttons, forms and small tools ^[1].

Who can use GPT-6 Intelligent UI? It rolled out to Plus, Pro, Business and Enterprise users from October 7 and to Free and Go users from October 8 ^[1].

Does generative UI replace designers? No. Someone still has to design the components, the rules and the guardrails the model works within, and research how people actually use the results ^[4].