Until now, ChatGPT was mostly one thing: a text window. Since October 7, 2026, OpenAI has been changing that at a basic level. With GPT-6 in Chat, the company is introducing “Intelligent UI” – answers that include charts, maps, forms, tappable buttons, and even small calculators right in the conversation. For millions of users, the question-and-answer machine becomes a tool that builds the right interface for itself.
Key takeaways
- GPT-6 in ChatGPT can combine text, graphics, and interactive elements in one answer and picks the format to fit the question.
- On request, ChatGPT builds small tools in the chat, such as a savings calculator, a bill splitter, or a simple game.
- Paying customers (Plus, Pro, Business, Enterprise) get GPT-6 Sol starting October 7; Free and Go get GPT-6 Luna starting October 8.
- The change applies only to the Chat experience; the models behind Work and Codex are not changing.
- OpenAI itself admits that the model’s design judgment still needs work.
What Intelligent UI actually does
The core of the update: OpenAI trained GPT-6 not just to write an answer but to design it. According to the announcement, the model decides on its own how text, images, and interactive elements fit together. A comparison then appears side by side, and an explanation becomes a diagram you can click through. When a plain text answer is the most useful one, it stays text.
The examples OpenAI shows are deliberately ordinary. If you are planning a Sunday roast for friends but do not yet know how many are coming, you get a guest-count slider next to the recipe that recalculates the shopping list, plus a cooking timeline as a checklist. On a road trip, the stops land on a map with notes on which detours are worth it. For learning, OpenAI points to interactive explanations of the central limit theorem, gross domestic product, drone photography, and the Monty Hall problem: change a value and immediately see what happens.
The third category is tools on demand. A calculator showing how retirement savings could grow, a bill splitter for dinner with friends, or a small retro game can, according to OpenAI, be created simply by asking in the chat. If that gets too busy, TechCrunch reports that users can dial back the number of visuals, much like other settings for ChatGPT’s personality.
How it works under the hood
The model is not writing arbitrary web pages. OpenAI built a library of ready-made, “streamable” components, meaning elements that can be displayed while the answer is still being generated. A compiler, a program that translates the model’s output, assembles the interface continuously as the model produces it. The answer therefore builds up step by step instead of appearing only once it is complete. The components keep the look consistent on the web and in the mobile apps, while the model decides how to arrange them.
For training, OpenAI says it evaluated the generated interfaces for clarity, usefulness, and completeness. That was meant to teach GPT-6 when interactivity helps and when a paragraph of text is enough. The closing caveat is remarkably candid: the model’s design judgment still needs work.
At the same time, OpenAI is speeding up responses. GPT-6 can begin answering while it is still thinking. For questions that need web search, the English original of the announcement says GPT-6 Instant starts answering 44 percent sooner on average than GPT-5.6 Instant; the German version of the same page shows 32 percent in a chart. Both are the vendor’s internal measurements, and independent measurements are still pending.
Who gets it and when
The rollout is global and staged. Plus, Pro, Business, and Enterprise have had GPT-6 with Intelligent UI in the Chat tab since October 7, powered by GPT-6 Sol. Starting October 8, the free tier and the low-cost Go plan follow, there with the GPT-6 Luna model. According to OpenAI, both are tuned for everyday conversation. In companies, availability depends on admin settings. If Sol and Luna are new to you: OpenAI introduced the first GPT-6 models for paying customers in September, as we reported at the launch of GPT-6 Sol.
One point matters for anyone using ChatGPT at work: the update only affects the regular chat. OpenAI says the models behind Work and Codex remain unchanged. If you have established workflows there, you do not need to adjust anything.
Analysis: software that adapts to the user
OpenAI frames the ambition broadly: instead of people learning how to use software, software should adapt to people. The practical benefit is already tangible, though. Many everyday questions are really small math problems – converting quantities, splitting costs, playing through scenarios. Until now, ChatGPT answered with a table or a paragraph you had to request again with every change. A slider that recalculates instantly removes exactly that loop.
With more capability, more responsibility shifts to the user. A nicely designed savings calculator looks more reliable than a paragraph of text, but it computes with the same assumptions the model made. Anyone using the numbers for real decisions should briefly check the underlying values. Especially for finance and health topics, the calculator remains an aid to understanding, not advice.
Strategically, the move is interesting as well. OpenAI is now bringing its new model generation to what it says are more than 1.2 billion people who use ChatGPT every week. Shortly before, the company announced that it would test visual ads in ChatGPT. An interface that becomes more visual and interactive is also a more attractive environment for such formats.
Outlook
Intelligent UI is the first major attempt to break up the rigid chat format, and it starts in the right place: with tasks that a slider, a map, or a checklist simply solves better than text. Whether the model reliably recognizes when a graphic helps and when it only distracts will only become clear in the daily use of hundreds of millions of people. Trying it is worthwhile as soon as the staged rollout reaches you, even without a subscription: a question you would normally have recalculated several times is the best test.