GPT-6 Intelligent UI: answers you can interact with
OpenAI’s ChatGPT update combines text, visuals, and controls around the task at hand.
Intelligent UI official release artwork. Credit: OpenAI. Original image ↗
OpenAI’s October 7 announcement introduces Intelligent UI in ChatGPT. Responses can combine ordinary text with diagrams, charts, buttons, forms, and interactive tools.
Ask for a task, inputs, and a result
An interactive answer is useful when changing an input helps you understand a decision. For a dinner bill, the inputs might be the total, tip percentage, and number of people; the result is each person’s share. For a lesson, the input might be a probability and the result a changing diagram.
A concrete request could be: “Create a bill splitter with editable total, tip percentage, and number of people. Show the calculation and each person’s share. Explain how rounding is handled.” This is an illustrative request, not a guarantee of a particular layout. It makes the intended behavior testable instead of asking only for something visual.
The announcement describes native components and a compiler that streams the interface as it is generated. The important product distinction is that controls can become part of the answer, rather than a screenshot of an imagined tool.
What changes for the reader
An explanation can expose controls for exploring a concept. A comparison can arrange choices side by side. A calculation can become a tool whose inputs you change within the conversation. The model chooses the format according to the question.
OpenAI describes a native component library and a compiler that streams the interface as it is generated. The announcement also describes responses beginning while the model continues reasoning.
Work through a calculator before trusting its controls
Consider the bill splitter suggested above. Let B be the bill in currency units, t the tip fraction, and n the number of people. An equal share is:
For an illustrative bill of 120, a 15% tip, and four people, S = 120 × 1.15 / 4 = 34.50. The graph changes the group size while holding the bill and tip fixed. The falling curve is a consequence of dividing a fixed total, not a measurement of the model's reasoning quality. A good interface should reveal this relationship rather than merely outputting a polished number.
Now change the bill to 100 and split a 115 total among three people. Rounding every share to 38.33 collects only 114.99. An exact payment plan could allocate 38.34 to one person and 38.33 to the other two. The sum remains 115. The interface must explain that rule; hiding the rounding discrepancy makes the apparent answer less useful. These amounts are arithmetic examples, not payment instructions or an assertion that a generated tool implements this policy.
Use boundaries as part of the specification
An input of zero people should not produce infinity as an ordinary result. A missing tip should have an explicit default, and the field should distinguish 15% from a fraction of 15. Testing these boundaries gives a concrete definition of correctness. The same approach transfers to a savings calculator, a recipe scaler, or a probability demonstration: identify valid inputs, compute a known result, then check how the tool handles invalid ones.
A generated chart also needs evidence appropriate to its purpose. A bill-splitting curve can use a disclosed formula; a factual population chart needs a dataset and a date. Controls should retain their labels, units, and keyboard behavior as the answer grows. A mobile layout should not change the meaning of the calculation by hiding an important input. Native, streamed UI makes interaction available inside a response, but the reader still needs to understand which information is a given input, a model assumption, or a calculated output.
Equal shares of a fixed 138 total
Focus the graph and use arrow keys, or tap a point, to inspect values. The complete data table is available below.
View data table
| People | Share (currency units) |
|---|---|
| 1 | 138 |
| 2 | 69 |
| 3 | 46 |
| 4 | 34.5 |
| 5 | 27.6 |
| 6 | 23 |
Availability has boundaries
The release applies to ChatGPT’s Chat experience. It uses GPT-6 Sol for paid tiers named in the announcement and GPT-6 Luna for Free and Go. It does not change the models powering Work and Codex as part of that release.
The current help article distinguishes supported thinking settings and notes that the Pro reasoning option does not support Intelligent UI.
Check the interaction before trusting the answer
Change one input at a time and compare the displayed result with a calculation you can inspect. Try an empty field, a zero value, and a value near a boundary. Check that labels explain units and that the controls remain usable on a phone.
For research or factual explanations, a polished chart still needs a source and an accurate scale. Ask what data the graphic uses and which assumptions the model supplied. For calculators, look for a visible formula or explanation of the computation.
Availability is tied to the ChatGPT Chat rollout and workplace settings described in the release. This article does not establish that the same interface is an exported API widget or a change to Work and Codex. Start in the documented product, test a small task, and review the response’s reasoning alongside its presentation.
