Short answer: ChatGPT Intelligent UI is GPT-6’s way of answering inside the Chat tab with a working interface instead of only paragraphs: a calculator you can change, a comparison you can read side by side, or a diagram you can explore. Open ChatGPT on the web, choose GPT-6 in Chat, pick a thinking level from Instant to Extra High, and ask for the tool by name, with its inputs, outputs, and rules stated plainly.
That last sentence is the whole method. Intelligent UI is not a separate app, a plugin, or a paid add-on. OpenAI introduced it with GPT-6 in Chat on 7 October 2026, rolling out to Free and Go users from 8 October 2026, according to OpenAI’s announcement. The model decides the format for each answer, so the skill is not learning a hidden menu. It is writing a request that names the result you need, the numbers you will change, and the checks the result must pass.
The trap is that an interactive result looks finished. A tidy calculator with a bold total feels more reliable than a paragraph, even when its formula is wrong. OpenAI itself says its design judgment is still improving. Treat every interactive answer as a draft with a nice front end: use it to explore, then verify the arithmetic before the result reaches a customer, a quote, or an invoice.
What Intelligent UI actually changes
Before GPT-6, a ChatGPT answer was text first. To see a quote at 30 guests instead of 20, you asked again and compared two blocks of text. Intelligent UI changes the unit of the answer: OpenAI says GPT-6 composes responses from text, visuals, and interactive elements, choosing how they fit together based on the question. The answer can now contain the tool, not just describe it.
OpenAI’s examples include a plan laid out so timing sits beside the recipe, a road-trip route on a map, an interactive diagram for a complex topic, and tools built on the spot such as a savings calculator, a bill splitter, or a game. A comparison can arrive side by side instead of as two lists. Text is not replaced: OpenAI is explicit that when a simple text answer is most useful, ChatGPT can still give you that.
Three boundaries matter. First, it lives in the Chat tab only. OpenAI’s help page states it is not available in the Work tab or in Voice, and the announcement says the models powering Work and Codex did not change in this release. Second, it needs GPT-6 at thinking levels from Instant through Extra High; Pro reasoning uses GPT-6 Astra and does not support it. Third, state is fragile: some components can retain state when you refresh a thread, but state is not retained across threads, so copy a tool’s inputs, formula, and checked results into your own document.
Who gets it, and on which model?
Access follows the GPT-6 rollout in Chat, not a separate subscription. OpenAI announced the rollout starting 7 October 2026 for Plus, Pro, Business, and Enterprise users, expanding to Free and Go users from 8 October 2026. Enterprise availability depends on workplace administrator settings. Those are announced start dates, not a promise that every account saw the feature that day, and staged rollouts mean some accounts still see it later.
| Plan | Announced rollout start | Model behind Chat |
|---|---|---|
| Plus, Pro, Business, Enterprise | 7 October 2026 | GPT-6 Sol |
| Free, Go | 8 October 2026 | GPT-6 Luna |
Both Sol and Luna support the feature, according to OpenAI’s help page, so you do not need a paid plan to try the prompts here. The difference is your plan’s normal usage limits, not a separate allowance. OpenAI announced no separate price for it. For context, Plus is $20 a month and Business is $20 per user a month on annual billing or $25 monthly with a two-seat minimum, per converging October 2026 pricing trackers, but check the pricing page inside your own account before buying. Do not buy a plan for this feature alone; try it on the plan you have, and upgrade only when you can name a limit you keep hitting.
How do you set up ChatGPT Intelligent UI in five steps?
There is no activation switch and no special command to memorise. Setup is a checklist that puts you in the right tab, on the right model, at a supported thinking level, with a request written well enough to build the right thing.
- Open ChatGPT on the web and start a new Chat. The web version is the most widely supported surface for this feature.
- Choose GPT-6 in the model menu. Look for GPT-6 itself, not Sol. An explicitly selected older model stays on that model.
- Pick a supported thinking level. Use Instant for a quick tool, Medium for a standard job, and High or Extra High when reasoning matters more than speed. Leave Pro reasoning unselected.
- Write the request as a brief, not a question. Name the tool, list every input with its starting value, list every output, state the formula in words, and state the edge-case rules. The prompt below shows the exact shape.
- Test the tool before you use it. Change one input, check that every dependent result moves, test zero and an invalid value, then restore the starting values. Only then copy checked results into your own document.
If the answer comes back as plain text, that is not necessarily a failure of your account. The model decides when interactivity helps. Ask once, in the same chat: “Show editable inputs and a working calculator inside this conversation, rather than only describing the calculations.” If it still returns text, run the access checks below. Do not count a static table, a generated image, or a downloadable web page as a working tool inside the conversation.
Your first working tool: a workshop cost calculator
The best first test is a calculator with few inputs and a result you can check by hand. The example below uses fictional figures, adapted from a tested walkthrough published by The Rundown on 8 October 2026. Keep it separate from real bookings and payments while you learn the pattern.
Copy and paste this prompt into a new GPT-6 Chat:
Create an interactive workshop cost calculator directly in this ChatGPT conversation. Use five editable inputs: 20 attendees, a $200 fixed room fee, $18 for food per attendee, $7 for materials per attendee, and a $900 total budget. These are fictional US-dollar figures. Total cost equals the room fee plus the number of attendees multiplied by food and materials per attendee. Show total cost, cost per attendee, and budget remaining. Update every result when an input changes, and label an amount over budget clearly. Keep the formula visible. Use whole, nonnegative attendee counts and nonnegative dollar amounts. Reject blank or invalid values with a clear message. At zero attendees, retain the fixed room fee and show cost per attendee as “Not applicable.” Start with no buffer, taxes, or other fees. Include a way to restore the starting values. Do not browse, connect apps, book anything, or send information elsewhere. If this chat cannot show an interactive calculator, say so rather than presenting a static picture as a working tool.
Check the starting values before you touch anything. The total should be $700, the cost per attendee $35, and the budget remaining $200, since $200 + 20 x ($18 + $7) = $700, which is our calculation from the prompt’s own figures. If those numbers do not match, fix the tool before trusting anything else it produces.
Now change only the attendee count from 20 to 30. Food and materials rise from $500 to $750, the total rises to $950, the cost per attendee falls to $31.67, and the budget remaining becomes negative $50, which the tool should label as over budget. This teaches the real lesson: a larger group can cost less per person and still break the budget, because the fixed room fee is shared while per-person costs keep climbing. Check the other inputs did not change, then restore the starting values and confirm the $700 total returns.
Run the edge cases next. At zero attendees, the total should stay at $200, the cost per attendee should read “Not applicable,” and the budget remaining should be $700. At 28 attendees, the total should be exactly $900, the budget. At 29, it should be $925, $25 over budget. A negative, fractional, or blank attendee count should produce a clear error, not a rounded headcount or a plausible total. If a result fails, send a repair prompt stating the correct figure and asking the model to explain the difference, fix the calculation, and recheck all four cases.
Then extend it with one change at a time: an editable budget buffer starting at 10% of the subtotal, added once. At 20 attendees, the final cost is $770 and the budget remaining $130; at 30, the final cost is $1,045, $145 over budget. Re-check every result that depends on the new rule, because a correct headline total can sit beside a stale per-person figure. Finally, ask for a short plain-text record with inputs, formula, results, and assumptions, and copy it into your own document. That record, not the chat thread, is what you keep.
Three more tools worth building this week
The same brief pattern transfers to other repeat jobs. A quote calculator takes hours, hourly rate, materials, and a discount or tax rule, and shows subtotal, discount, tax, and final total with a visible formula; check the starting quote by hand and state the rounding rule, because a rounding difference from your invoicing system is a real cost. A side-by-side comparison pastes two or three options and shows price, capacity, limits, and missing details in the same fields, with every unknown labelled rather than filled in; do not ask for a winner until the required fields have support. An explainer or quiz for training works the same way: a diagram with selectable parts, or a five-question quiz built only from notes you paste in, flagging questions the notes cannot settle. For prompt structure, our guide on writing better ChatGPT prompts covers the same discipline, and our walkthrough of the ChatGPT Chrome extension shows how Chat differs from an agentic browser mode: this feature presents an answer, it does not book, pay, or send anything by itself.
Try the maths yourself: a workshop cost calculator
Use the calculator below to rehearse the checks this article asks you to run inside ChatGPT. Change the attendee count and watch the per-person cost fall while the total rises, then try zero attendees and an invalid value. The figures are examples only, not a quote for any real venue.
How this calculator works: subtotal = room fee + attendees x (food + materials); buffer = subtotal x buffer percentage / 100; total = subtotal + buffer; cost per attendee = total / attendees, shown as "Not applicable" at zero attendees; budget remaining = budget - total. At the starting values with no buffer, that gives $200 + 20 x ($18 + $7) = $700, $35 per attendee, and $200 remaining, which is our calculation from the example figures. At 30 attendees, the total is $950 and the budget remaining is negative $50, which is why the status reads as over budget.
What to check before you trust any interactive answer
Interactivity is not verification. A tool that updates instantly can still carry a wrong formula, a missing fee, or a silently rounded input, and polish makes the error harder to spot than a wrong sentence would be. Apply these checks before the result leaves your hands.
- Check the starting values first. Recompute the headline total by hand. If it is wrong, nothing downstream is trustworthy.
- Change one input at a time. Confirm every result that should move actually moves, and untouched inputs stay unchanged.
- Test zero, the budget boundary, and an invalid value. Zero should not divide by zero. The exact budget point should read as exactly on budget. An invalid value should produce an error, not a plausible total.
- Read the formula, not just the total. Ask for the formula to stay visible. Tax, fees, discounts, and cancellation terms are the usual missing pieces.
- Keep the record outside the thread. Ask for a plain-text summary with inputs, formula, results, and assumptions, and save it in your own document.
The same discipline applies to pricing. Our comparison of ChatGPT API versus Plus and Business plans makes the point from the pricing side: the headline plan is rarely the whole cost, and the billing unit decides the bill. Inside a single answer, the headline total is rarely the whole story, and the formula decides whether the total means anything.
Why you might not see it, and what to do
Most "it does not work" reports turn out to be one of a handful of mismatches. Work through them in order rather than rewriting the prompt five times.
| Check | What to do |
|---|---|
| Tab and model | Use the Chat tab with GPT-6 selected. It is not in the Work tab or Voice, and an older model will not show it. |
| Thinking level | Pick Instant through Extra High. Pro reasoning uses GPT-6 Astra and does not support it. |
| App surface | Try ChatGPT on the web or an updated app. Older desktop apps for Mac and Windows are reported not to support it. |
| Rollout and workspace | Allow for staged rollout, and check administrator settings on Business and Enterprise accounts. |
| Prompt shape | Ask for the tool by name, with inputs, outputs, formula, and edge-case rules. A vague question may correctly return plain text. |
| Too much visual noise | On the web, go to Personalization, then Layout and Visuals, then Simple. OpenAI says this reduces visual elements, though some may still appear. |
If you prefer mostly plain answers, that Simple setting is the honest off-ramp. It does not promise a fully text-only experience. Custom Instructions can steer the style, but OpenAI's help page warns that preferences guide the response and the exact layout can vary. Treat a preference as a nudge, not a guarantee.
Who should not build on it yet
This feature is a presentation layer, not a system of record, and some jobs should stay outside it. If the result must be auditable, shared across a team, or reopened months later, build it in a spreadsheet or a real tool and use ChatGPT only to draft the logic. If the job involves booking, payment, or messaging a customer, do not route it through a chat interface never presented as an action layer. If the numbers carry legal or tax consequences, a human who understands the rule must sign them off. The same caution applies to data in a test chat: our guide to setting up the Gemini business hub makes the point about connecting a store or accounting app before checking what the connection can read and change, and here the tool asks for your numbers in the chat itself, so paste rounded, non-sensitive figures and keep account numbers and customer data out entirely. Finally, set the decision before you open the chat: what number would change your mind? Without a stopping rule, another tweak is just a slower way to avoid deciding.
What we cannot confirm
Some points here rest on secondary reporting, labelled as such. The claim that older desktop apps do not support the feature comes from secondary guides summarising rollout notes, not from the help page text we opened, which names no desktop versions. The Plus and Business prices come from converging October 2026 pricing trackers, not from a pricing page opened during this run, so check them inside your own account before purchasing. One secondary guide, at OmniaKey's GPT-6 guide, notes the launch and help pages are not fully synchronised; we did not observe that in the help text we opened, which names GPT-6 Luna for Free and Go, so treat that report as unconfirmed.
What OpenAI's own pages do confirm is the core of this guide: the feature lets GPT-6 combine text, visuals, and interactive elements in Chat; it works from Instant through Extra High; Pro reasoning with GPT-6 Astra does not support it; it is not in the Work tab or Voice; and the Simple setting reduces, but does not guarantee the removal of, visual elements. For model choice, see our guide on comparing AI models on your own work, our overview of AI tools for small business owners, and our breakdown of the real cost of AI tools.
The takeaway
Start with one real job, one prompt written as a brief, and one hand-check of the starting total. Build the workshop calculator, change the headcount, test zero and the budget boundary, and save the plain-text record. Then transfer the pattern to a quote, a comparison, or a training explainer, keeping the same checking habit every time. The model will keep getting better at choosing formats. Your job is to keep getting better at deciding whether the format earned your trust.
Frequently asked questions
What is ChatGPT Intelligent UI?
It is GPT-6's ability in the Chat tab to answer with text plus visuals and interactive elements, such as calculators, comparison cards, diagrams, forms, and tappable controls, instead of paragraphs alone. OpenAI introduced it with GPT-6 in Chat on 7 October 2026.
Do I need a paid plan to use it?
No. OpenAI rolled it out to Plus, Pro, Business, and Enterprise users from 7 October 2026 and to Free and Go users from 8 October 2026, with GPT-6 Sol on paid plans and GPT-6 Luna on Free and Go. Your normal plan limits still apply, and availability can arrive in stages.
Why does my answer come back as plain text?
The model decides when interactivity helps, and a short question may be best answered in words. Ask once in the same chat for editable inputs and a working tool inside the conversation. If it still returns text, check that you are in the Chat tab with GPT-6 selected at a thinking level from Instant to Extra High, and try the web version.
Does it work in the Work tab or Voice?
No. OpenAI's help page states it is available in Chat, not in the Work tab, and not in Voice. It also does not apply to the models powering Work and Codex in this release.
Why does Pro reasoning not show interactive answers?
The Pro reasoning option uses GPT-6 Astra, which does not support it. A Pro subscriber must choose a supported thinking level, from Instant through Extra High, to get the interactive format. A Pro subscription and Pro reasoning are different settings.
Can I turn it off?
Partly. On the web, go to Personalization, then Layout and Visuals, then Simple. OpenAI says this reduces visual and interactive elements, though some may still appear. You can also ask for a plain-text answer in an individual conversation.
Shurah is the founder of AI Tools Daily, tracking pricing, licensing and policy changes across AI tools so readers can make decisions without wading through marketing claims themselves.
