GPT-6 in ChatGPT can now answer some questions with an interactive interface instead of text alone. OpenAI calls this Intelligent UI: an answer may contain a chart, controls, a form or a small tool you can use inside the conversation. The useful habit is to check what the interface assumes before acting on its result. A slider can make a wrong calculation feel more convincing because it moves smoothly. OpenAI’s October 7 announcement
Source checked October 8, 2026. This is a documentation-based explainer, not an AIViewer product test. The exercise below is fictional teaching material.
What changed in ChatGPT?
OpenAI says GPT-6 can choose a response format to suit a question. Its examples include a visual explanation, an adjustable shopping list and calculator, and a bill splitter. A straightforward question can still receive a straightforward text answer. The company says the interface can appear progressively while the response is being generated. These are provider descriptions; AIViewer has not tested how consistently they work in an account. OpenAI’s product description
This release concerns the Chat tab in ChatGPT. OpenAI says it does not change the models powering Work or Codex. GPT-6 with Intelligent UI began rolling out on October 7 to Plus, Pro, Business and Enterprise; Free and Go rollout was scheduled to start October 8. OpenAI identifies GPT-6 Sol for the first group and GPT-6 Luna for Free and Go. Enterprise access depends on administrator settings. The separate Pro reasoning option uses GPT-6 Astra and does not support Intelligent UI. A rollout announcement does not prove the feature is present in your account today. OpenAI’s availability section OpenAI’s feature help
If you prefer fewer visual elements, OpenAI documents Personalization → Layout and Visuals → Simple in ChatGPT web settings. It says this reduces interactive elements, although some may still appear. Asking for a different response style in an individual conversation is another option; the exact layout can vary. OpenAI’s feature help
OpenAI also describes answers that can begin while GPT-6 continues reasoning. Its published speed and quality comparisons are internal evaluations, not independent AIViewer measurements. They do not tell you whether a particular generated calculator has the correct inputs or formula. OpenAI’s evaluation discussion
A checkable example: workshop seats
The following case is fictional teaching material. The example interface and its possible error were written for this lesson; they are not screenshots, transcripts or observed GPT-6 results.
A community group has a room with 30 total seats. It has 24 confirmed registrations. It is holding 2 separate seats for volunteers; those two people are not included among the 24 registrations. Five people are on a waitlist, but none has been offered a seat. The group wants to know how many seats are available for new confirmed registrations.
An answer must use the group’s actual rule:
Available seats = total seats − confirmed registrations − seats held separately
= 30 − 24 − 2
= 4The waitlist does not reduce availability until someone is offered and accepts a seat. If the two volunteer seats were already counted in the 24 registrations, subtracting them again would be another error. Define the categories before doing the math.
If an interactive answer says 6 seats available, it may have calculated 30 − 24 and overlooked the two seats held separately. The display might be neat and responsive, but the result would still be wrong for this source. That is a deliberately flawed authored example, not a claim that GPT-6 made this mistake.
Ask for a result you can inspect
If Intelligent UI is available in your account, try a request like this using only the fictional figures above:
Make a small seat-availability calculator for this fictional workshop.
The room has 30 total seats, 24 confirmed registrations, and 2 seats
held separately for volunteers. The volunteer seats are not included
in the confirmed registration count. Five people are on a waitlist,
but none has accepted an offer.
Show every input and the formula beside the result. Let me change the
number of confirmed registrations. Keep total seats and separately
held seats visible. If the result would be below zero, label it as
over capacity. Do not treat the waitlist as confirmed registrations.This is a request for an inspectable draft, not a guarantee that the generated interface will follow every rule. If your account does not offer Intelligent UI, the same prompt can produce a table or written calculation that you check in the same way.
| Check | What the source says | Correct treatment |
|---|---|---|
| Total seats | 30 | The room limit, including the two held seats. |
| Confirmed registrations | 24 | Count only accepted registrations. |
| Separately held seats | 2 | Subtract them once because they are outside the 24. |
| Waitlist | 5 | Show separately; do not subtract from available seats. |
| Current result | 30 − 24 − 2 | 4 seats available. |
Before you use a generated tool with real information, inspect the labels as carefully as the answer. “People interested,” “registered,” “attending” and “seats held” can mean different things. A correct formula applied to the wrong category is still a wrong result.
Change one input and check again
Move the hypothetical confirmed-registration value from 24 to 27, leaving the other values unchanged. The correct result is 30 − 27 − 2 = 1 available seat. If the interface still shows 4, its control is not connected to the result. If it shows 3, it has again ignored the held seats.
Try it yourself: suppose 31 people are confirmed while the two separate volunteer seats remain held. What should the interface show? Answer: 30 − 31 − 2 = −3. It should show over capacity by 3 seats, not “−3 available seats” as if that were a usable inventory. The label matters as much as the arithmetic.
For a real event, first check the current registration record, the room’s actual permitted capacity and whether held seats overlap with registrations. The exercise checks fidelity to the stated inputs; it does not validate a venue record. Review any generated output before sharing it or changing a registration system.
What to take from the release
Intelligent UI could make explanations and small calculations easier to explore. Its most useful first test is a task with a known answer and a value you can change. Ask to see the inputs and rule, calculate one result independently, then change a single input and repeat the check. That gives you evidence about this output in your account without assuming the new interface is reliable for every task.
For a broader method of checking factual claims and sources, use our guide to verifying an AI answer. For document summarization where a polished output can also hide changed numbers, see how to summarize a report without changing the facts.