For two years, ChatGPT answered questions with text. On Oct. 7, OpenAI began pushing out a version that answers with interfaces.
The company rolled out GPT-6 to every tier of ChatGPT, and with it a feature called Intelligent UI that turns the model’s responses from plain text into something a user can act on. Charts can be edited in place. Buttons can be clicked. The model can generate a working calculator for a specific task on the spot, rather than describing how the math might be done.
The rollout began with the four paid tiers, with free and Go users following the next day. Users can dial back the visual elements if they prefer a plainer answer. The design choice is the point: OpenAI is betting that the best response to many questions is not a paragraph but a small application.
“In most scenarios, the optimal answer is not just text,” Aarush Selvan, an OpenAI product manager, said at a press briefing, according to reports. TechCrunch and The Verge both described the release as a step in ChatGPT’s shift from a question-and-answer tool into something that gets things done.
The distinction matters to OpenAI’s business. A chatbot that only talks is paid for by subscription. A system that can complete a task, book a trip, file a form or adjust a chart is closer to a platform, and platforms command a different kind of revenue and a different kind of loyalty.
The move also puts ChatGPT more directly in the path of the software people already use. When a model can render a spreadsheet, a calendar view or a calculator inside its own answer, it competes with the tools those interfaces were built for. OpenAI is betting that users would rather stay inside one assistant than switch between applications.
The shift has been building for months. ChatGPT began as a chat window, then added search, then file analysis, then the ability to write and run code. Each step moved the product closer to doing rather than saying. Intelligent UI is the point where that line of development becomes visible in the answer itself, instead of in the toolbars around it.
The pricing structure tells part of the story. OpenAI has kept the strongest capabilities in its paid tiers, and releasing GPT-6 across all of them at once suggests the company wants paying users to feel the new model before it reaches the free product. The speed of the rollout is unusual for a release this large, and it signals how much competitive pressure OpenAI is feeling.
The accuracy question will not go away. Interactive elements look authoritative in a way that plain text does not, and a model that renders a chart wrong has produced a more confident-looking mistake. OpenAI has said users can reduce the visual elements, but the default now invites the model to do more, which means there is more to get wrong.
The shift from text to interfaces is not confined to OpenAI. Across the industry, assistants are learning to act, and the companies that build them are betting that the market will reward doing over saying. Whether users follow is the experiment now running at scale inside every ChatGPT window.
The competitive pressure points the same way. Google and Anthropic have been racing to turn their own assistants into agents, and OpenAI cannot afford to let the market’s expectation of what an assistant does move ahead of what ChatGPT delivers. Releasing GPT-6 across every tier at once, rather than in stages, is a statement that this is the product’s new baseline.
The risk is familiar, and it is specific to interfaces. A paragraph with a mistaken number is easy to question. A button that silently does the wrong thing is not. The more a model does on a user’s behalf, the higher the cost of its errors, and the harder it becomes to inspect what it actually did.
OpenAI has been pushing for months toward so-called agents, systems that take actions on a user’s behalf rather than merely drafting text. Intelligent UI is the visible edge of that push, the part a user sees and touches. The underlying reasoning may be opaque, but the output is now concrete enough to interact with.
The bet is that people would rather do than read. The answer to “how much will this cost” is more useful as a calculator than as a sentence, and the answer to “what would this chart look like” is more useful as a chart. If OpenAI is right, the chatbot’s days as a plain text box may be numbered.
What is less clear is who controls the results. An assistant that can act is an assistant that can be blamed, and OpenAI is now asking users to trust its model to answer correctly and then to do correctly. That is a larger promise, and the company has yet to show it can keep it at the scale this release invites.
