OpenAI Begins GPT-6 Rollout With Intelligent UI in ChatGPT

OpenAI has begun rolling out GPT-6 in ChatGPT’s Chat tab, bringing with it a new capability called Intelligent UI that can combine written answers with visuals and interactive elements. The company says the update is initially rolling out globally to ChatGPT Plus, Pro, Business and Enterprise tiers, with Free and Go users scheduled to receive it starting tomorrow.

Intelligent UI is designed to change how ChatGPT presents an answer, rather than simply adding more visual decoration to an existing text response. Depending on the question, the system can generate graphics, tappable buttons, forms, charts and other interactive experiences that users can work with inside the conversation.

A move beyond text-only answers

OpenAI says ChatGPT has “predominantly been a text-based interface”, but argues that text is not always the most useful format for planning, learning or completing an everyday task. Intelligent UI allows GPT-6 to decide whether a question is best answered through text, a visual explanation, an interactive diagram, a comparison or a tool built for the task.

A comparison, for example, can be presented side by side, while an explanation may be easier to follow through an interactive diagram. ChatGPT can still provide a conventional written response when that is the clearest or most useful option.

OpenAI says GPT-6 has been trained to compose responses using text, visuals and interactive elements, and to choose how those elements should fit together based on the user’s question. The company’s stated aim is to make everyday answers more visual and to make complex topics easier to learn.

The visual elements are also intended to be usable rather than merely illustrative. Instead of receiving only a static image, users may see tappable buttons, customised calculators, interactive charts, editable graphs and other components that respond to what they do in the conversation.

The amount of visual content can be adjusted. Search Engine Journal reported that users who prefer fewer visuals can turn off the “Layout and visuals” setting on the web, although some visual elements may still appear.

TechCrunch reported that OpenAI did not describe the feature as a way to make every answer graphical. The company’s product manager Aarush Selvan said the more important goal was to help users “get things done”, including planning a recipe, planning a trip or trying to learn something.

Diagrams for learning and explanation

One of the demonstrations described by TechCrunch involved a college student trying to understand how an airplane wing generates lift. After the request to “explore how an airplane wing generates lift”, ChatGPT produced a set of diagrams intended to show how planes stay in the air.

The example illustrates the distinction OpenAI is drawing between an answer that describes a concept and one that lets a user explore the concept visually. The company says visual and interactive explanations can allow users to change an input, see what happens or work through a problem step by step.

A bicycle example explains how a 7-speed bicycle brings together a frame, handlebars, wheels and hand-operated brakes. It also describes how seven gears allow a rider to adjust pedalling effort as the drivetrain turns effort into motion.

The bicycle example is one of the ways GPT-6 can use an interactive or visual explanation for a familiar mechanical system. The broader promise is that the format of the response can follow the structure of the question, rather than forcing every subject into the same sequence of chat messages.

OpenAI says this can include a diagram for a scientific concept, a visual recipe, a map for a multi-day hiking excursion or an interactive explanation of bicycle mechanics. The examples point to different kinds of use: diagrams for abstract ideas, maps for journeys, and visual instructions for practical activities.

The feature is also intended to support direct exploration. A user can ask ChatGPT to create a calculator for examining how savings could grow, a bill splitter for dinner with friends or a game that can be played within the conversation.

Those tools do not require the user to describe the task and then leave the conversation to find a separate page. Search Engine Journal said the calculator and bill-splitting examples use generic inputs and public mathematics.

From a question to a working tool

OpenAI describes Intelligent UI as a system that can combine user-interface elements, data and actions in response to a request. That means the output can be more than an explanation of what to do: it can include an interface for carrying out part of the task.

The company’s examples include a savings calculator, a bill splitter and a game. The savings calculator can be used to explore how savings could grow, while the bill splitter is intended for dividing the cost of dinner with friends.

Search Engine Journal noted that ChatGPT already offered interactive charts, which were added in June, and interactive modules for more than 70 mathematics and science topics, which were added in March, according to OpenAI’s release notes. Intelligent UI extends that direction by allowing GPT-6 to decide how an interface should be composed for the question being asked.

The change could put pressure on simple calculator and converter pages, according to Search Engine Journal, because users can ask ChatGPT to create and use a tool directly in the chat. The report said such tools may reduce the need to leave the conversation for some tasks.

OpenAI has not said in the announcement how sources will appear when an answer is presented as a chart, comparison or tool. The company’s search help page says answers that use web-search results may include citations, which can be opened through a Sources button when it is available.

That leaves a distinction between the interface used to explore an answer and the information used to support it. The announcement describes the new formats and their interaction model, but does not specify a universal source display for every chart, comparison or generated tool.

A roast plan shows the range of formats

The material describing Intelligent UI also uses the planning of a Sunday roast to show how ChatGPT can organise information around a task. In that example, the timing for preparing the meal can sit alongside the recipe, allowing the user to see how the steps fit together.

The suggested main dish is a garlic, rosemary and Dijon roast leg of lamb. The sides include crispy roast potatoes, roasted carrots, a large green salad, and peas with butter and mint.

The meal also includes proper lamb gravy, a quick mint sauce and apple crumble with vanilla ice cream. For drinks, the example recommends a medium or full-bodied red such as Rioja, Syrah or Cabernet.

The example describes a bone-in leg as suitable for having friends over because it is festive, relatively forgiving and easy to carve. It also says that if the group goes beyond about 8 people, two smaller legs may be easier to cook evenly than one very large leg.

The two legs can also be cooked to different levels, with one left slightly pinker than the other. The night before roasting, the example recommends seasoning the lamb with salt and rubbing it with olive oil, Dijon, garlic, chopped rosemary, black pepper and lemon zest before refrigerating it overnight.

On Sunday, the lamb should be taken out roughly an hour before roasting, according to the example. It recommends roasting at 425°F for about 20 minutes and then reducing the temperature to 350°F until the centre reaches roughly 125, 130°F for medium-rare or 135°F for medium.

The example says a thermometer is more reliable than calculating minutes per pound. It recommends resting the lamb for 30, 40 minutes under loose cover, a pause that also leaves the oven available for crisping the potatoes and finishing the vegetables.

For the potatoes, the suggested options are Yukon Golds or russets. They are peeled, cut into large chunks, parboiled for about 8, 10 minutes, drained and shaken in the pot to roughen their edges.

The potatoes are then roasted in hot olive oil or duck fat with salt, rosemary and whole garlic cloves. The stated goal is for them to be deeply golden and aggressively crispy rather than merely cooked.

The planning example assumes a meal at 6:00 pm in one part of the schedule. It says the main advantage of the arrangement is that almost nothing needs attention while friends are arriving.

For an undecided guest list somewhere between 6 and 10 people, the example suggests shopping for 8 and keeping flexible ingredients generous. It also says roast-lamb leftovers are useful, making it preferable to have slightly too much rather than discover that the meal is short.

The menu is described as a British-inspired Sunday roast with Mediterranean elements. Its combination includes crispy potatoes, tender rosemary-and-garlic lamb, seasonal vegetables and gravy, with the kind of meal where people remain at the table for another glass of wine.

The wider approach is to build a flexible menu that changes with the headcount. It includes a classic leg of lamb with lemon, garlic, rosemary and olive oil, served with homemade gravy and mint sauce.

The potatoes are parboiled, shaken to create fluffy edges and roasted until golden. The example calls that step “Nonnegotiable!”

It also proposes vegetables tossed with olive oil, honey and fresh thyme, along with Tenderstem broccoli with toasted almonds and lemon zest. A fresh element is included to balance the richness of the lamb, and the dessert is served with vanilla ice cream.

The dessert can be made the day before and reheated while the meal is being eaten. The shopping quantities are intended to be adjustable according to the guest count, with the assumption that everyone eats lamb and that the host is included in the total.

The portions are described as generous, with a little extra. The lamb weight includes the bone, and the example advises asking a butcher for help in selecting the closest available size.

The suggested shopping list also includes garlic, rosemary, thyme, lemons, honey, almonds, olive oil, gravy ingredients, mint sauce, crumble topping and vanilla ice cream. Another version of the schedule assumes that the meal is being eaten at 5 PM.

A separate example uses a roughly 2.4 kg bone-in leg of lamb for six people. It says the roasting time should be adjusted to the actual joint.

For food safety, the USDA recommendation cited in the material is to cook whole cuts of lamb to an internal temperature of 145°F, followed by at least a three-minute rest. The same material says actual cooking time varies according to the size of the joint and the oven.

It adds a simple starter of olives, good bread and whipped feta. The stated purpose is to give guests something to nibble while the cooking is finished without spoiling their appetites.

The roast example matters because it shows how the system can arrange a plan, quantities, timings, preparation steps and contingencies around one request. It is not simply a recipe in a different visual style; the response can connect the guest count, shopping list, cooking sequence and timing.

How Intelligent UI is generated

OpenAI says it built a library of native, streamable components for Intelligent UI. It also built a compiler that processes the interface while the model generates it.

The component library gives each response a familiar design foundation, while allowing the model to decide how the individual pieces should fit together. The compiler allows the interface to appear progressively instead of making the user wait for the entire response to be completed.

OpenAI says it expanded its training methods so GPT-6 could make decisions about content, layout, visuals and interaction. The company evaluated the interfaces created by the model for clarity, usefulness and completeness.

GPT-6 learned to use the component library and to make decisions about how information should be organised. Its training also covers when to use interactivity and when a simple text response is sufficient.

The model therefore has some discretion over the structure of the result. A user asking for a comparison may receive a side-by-side presentation, while someone seeking an explanation may receive an interactive diagram.

OpenAI says there is still work to do to improve GPT-6’s design judgment and expand what it can create. Even so, the company describes Intelligent UI as a significant shift in how people interact with ChatGPT.

The release is designed to work across web and mobile while remaining fast and familiar. OpenAI says the native components and progressive compilation are intended to support that experience as the model builds an answer.

Rollout across ChatGPT plans

GPT-6 with Intelligent UI is rolling out in the Chat tab for ChatGPT Plus, Pro, Business and Enterprise tiers. OpenAI says the rollout starts globally for those tiers, while Free and Go users begin receiving the feature tomorrow.

Enterprise availability depends on workplace administrator settings. The models used for ChatGPT Work and Codex are not changing as part of this release.

GPT-6 in ChatGPT is powered by GPT-6 Sol for Plus, Pro, Business and Enterprise users. Free and Go users receive GPT-6 Luna, with both models tuned for everyday conversation.

Search Engine Journal reported that GPT-6 replaces the “Latest” option in the model menu, while GPT-5.6 Sol remains available as a separate option. The Intelligent UI feature is being introduced gradually on the web and on supported, updated apps.

The feature is available at thinking levels from Instant to Extra High, according to Search Engine Journal. The Pro thinking level continues to run on GPT-6 Astra but does not support Intelligent UI.

The older ChatGPT desktop applications for macOS and Windows also do not support the feature. Sol and Luna launched on Sept. 22 for ChatGPT Work, Codex and the API, when OpenAI said they were not yet available in Chat.

The release therefore changes the Chat experience without changing the models that power Work and Codex. It also does not make Intelligent UI universal across every ChatGPT application or every thinking level.

Answers can begin while GPT-6 is still working

Intelligent UI is being introduced alongside a change in how GPT-6 delivers answers. OpenAI says ChatGPT can now begin answering while it continues to think.

With earlier reasoning models, users had to wait for the model to finish thinking before receiving an answer. GPT-6 is trained to interleave thinking with answering and to take the user’s waiting time into account.

The model can progressively compose a response using the knowledge and findings available at that point. OpenAI says it learned to build answers across multiple partial responses, with each response adding useful information without unnecessary filler.

The company also says the complete answer should remain cohesive and factual, rather than appearing as disconnected fragments. GPT-6 can share more insights while it is still thinking or working with tools, without waiting for the user to ask again.

This approach is connected to Intelligent UI because the interface can appear progressively as the model generates it. Users may therefore see both the answer and parts of the interface develop before the entire response is complete.

OpenAI’s reported performance gains

In an internal evaluation of high-value, everyday agentic tasks, OpenAI says GPT-6 Extra High began answering in the same amount of time as GPT-5.6 Medium. The company also says GPT-6 Extra High achieved a better overall score than GPT-5.6 Extra High in that evaluation.

For questions requiring web search, OpenAI says GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant. The company also says GPT-6 improves the quality of answers that require web search.

OpenAI says GPT-6 makes better decisions about when to look something up. It also says the model more reliably finds information that supports its answer.

In an internal evaluation of difficult problems, GPT-6 addressed the key aspect of the user’s question more often than GPT-5.6, according to OpenAI. The company did not provide a figure for that comparison.

These performance claims come from internal evaluations described by OpenAI. The material does not provide a broader independent test result for the new interface.

Safety and limits

OpenAI says the GPT-6 release builds on several of Astra’s safety advances. The stated goals include better resistance to attempts to bypass safety training, stronger adherence to safeguards and clearer communication about what the model can and cannot do relative to GPT-5.6 Sol.

The company updated GPT-6’s safety training using lessons from real-world use. The update is intended to strengthen protections against high-risk misuse involving cyberattacks, biological threats and violence.

In adversarial testing, OpenAI says GPT-6 showed stronger resistance to attempts to bypass its safety training, particularly attacks that adapt over multiple turns. The model is also trained to respond safely in higher-risk situations while avoiding unnecessary refusals of harmless requests.

GPT-6 uses conversation history and relevant context to recognise risks that may not be apparent from a single prompt, according to OpenAI. That approach is part of the company’s effort to make the model’s safety responses depend on more than one isolated request.

OpenAI says it conducted alignment evaluations throughout training. Those evaluations assessed honesty and the model’s responses to safety restrictions.

The company says GPT-6 is better at recognising and communicating the limits of its own capabilities. That includes acknowledging when it lacks the information or tools needed to answer.

More detail on the results and OpenAI’s review is available in the system card referenced in the announcement. The announcement presents the safety improvements as part of the wider effort to bring more capable AI to more people while improving how it recognises real-world risks.

A different model for software interaction

OpenAI frames Intelligent UI as a step towards a ChatGPT that can shape an interface around what a person is trying to do. The company says its longer-term aim is for ChatGPT to create more of the interfaces people need for everyday tasks.

The idea reverses the traditional relationship between people and software described by OpenAI. For decades, people have had to learn how to use software and navigate fixed interfaces designed around particular tasks.

OpenAI says the future could instead involve people describing what they want to accomplish and working with an experience built around that goal. Its stated belief is that software should adapt to people rather than requiring people to adapt to software.

The release brings that idea into the ChatGPT conversation through interfaces that can include data, controls, charts, forms and actions. It also keeps plain text available when no richer format is needed.

OpenAI says GPT-6 brings its latest advances in intelligence to the more than 1.2 billion people who use ChatGPT every week. The company presents Intelligent UI as an early step towards a system that can shape the response and the interface around the task itself.

TechCrunch reported that the feature may make ChatGPT friendlier and more accessible and could help bring in a wider audience, although OpenAI did not state those points directly. The company’s direct emphasis is on making complex topics easier to learn and helping users complete tasks where a text-only answer is not the most useful response.

That distinction is central to the release. GPT-6 is not only being presented as a model that can produce better prose, but as one that can choose whether an explanation should be written, drawn, mapped, calculated or turned into an interactive tool.

The company says there is still work ahead on design judgment and on expanding what Intelligent UI can create. For now, the rollout establishes the direction: ChatGPT can respond in different formats within the same conversation, while retaining text when text is the better answer.