Keep going
Keep learning with Rundown Pro
Pro adds the full on-demand course library, certifications, group office hours, and eligible member perks.
$29/mo, cancel anytime
Guides OpenAI published oct 8, 2026
ChatGPT’s new Intelligent UI can put calculators, diagrams, comparison cards, and other controls inside an answer. You can change an input or explore part of an explanation without asking a new question each time. OpenAI introduced the feature with its October 7 GPT-6 rollout in Chat.
In this guide, you will ask for a workshop cost calculator, change the number of attendees, and check whether the result updates correctly. Then you will add a budget buffer with a follow-up request.
The exercise uses fictional figures and needs no downloads, connected accounts, or coding. Keep it separate from real bookings and payments.
We tested the walkthrough in ChatGPT on the web with GPT-6 and Medium thinking on October 8, 2026. The starting calculations, changed inputs, reset, zero-attendee rule, and invalid-input checks passed. We also tested the buffer revision and checked its text summary. We asked for a compact layout after the first calculator; your layout may differ.
| Account | Announced rollout start | Model behind the new Chat experience |
|---|---|---|
| Plus and Pro | October 7, 2026 | GPT-6 Sol |
| Business and Enterprise | October 7, 2026, subject to workspace access | GPT-6 Sol |
| Free and Go | October 8, 2026 | GPT-6 Luna |
These are announced rollout start dates, not a promise of access on every account that day. OpenAI’s dated release notes describe this change in the Chat tab. The release leaves the models in Work and Codex unchanged. We did not test Free or Go access.
A Pro subscription and Pro reasoning are different. The Pro reasoning option uses Astra and does not support Intelligent UI. A Pro subscriber can use the supported GPT-6 thinking levels instead. Work accounts also depend on Business model settings or Enterprise permissions.
Choose a format that helps you do something with the answer. OpenAI’s Intelligent UI examples let readers compare, explore, calculate, and play within the conversation.
| What you need | What to ask for |
|---|---|
| Try several sets of numbers. | A calculator with editable inputs and visible formulas. |
| Compare options against the same needs. | Side-by-side cards using the same fields and source links. |
| Understand how several parts fit together. | A diagram with parts you can select for an explanation. |
| Learn through practice. | A short quiz that explains each answer after you respond. |
Start with one task. Ask for the result you need, the inputs you will change, and the rules it must follow. Leave decorative graphics out when they add little.
Compare these three workshop venues using only the details I paste below. Show the price, room capacity, accessibility information, and missing details side by side. Label every unknown. Do not choose a winner until the required fields have support.
[Paste the approved venue details and requirements.]Useful interactions should make the comparison easier to inspect. They cannot supply facts missing from the brief.
Explain what happens when I open a website. Use a simple diagram with parts I can select for a short explanation. Keep it to six steps, define new terms, and flag where you simplified the process.Check the explanation as well as the labels. Select each part and make sure the text matches the part you chose.
Use these training notes to create a five-question practice quiz. Show one question at a time, wait for my answer, then explain the answer using a quote from the notes. Keep a score and flag questions the notes cannot settle.
[Paste approved training notes.]These are suggested follow-up tasks, not tested demonstrations. Keep source material narrow enough to review each answer.
Check these before rewriting your prompt:
| Check | What to do |
|---|---|
| Rollout | Compare your account with the dated rollout table above. Access can arrive in stages. |
| Chat and model | Use GPT-6 in Chat for this exercise. |
| Thinking | Choose a supported level from Instant through Extra High where your plan offers it. |
| App | Try the web. Update mobile apps; older Mac and Windows desktop apps lack support. |
| Workspace | Ask an admin to check model access when a work account lacks the option. |
OpenAI documents the supported clients and controls, launch timing, and Enterprise permissions. A more specific prompt can help describe the result. It cannot unlock a feature your account lacks.
A useful AI tool needs correct rules and results, as well as clear controls. Once the small example works, test it with new inputs before relying on it for a recurring task.
Nate Grahek’s Intro to AI Evals covers test inputs, expected answers, and comparisons of quality, speed, and cost. Its examples use a spreadsheet and Claude Code. It teaches the testing method rather than this new ChatGPT interface.
Use ChatGPT on the web for this walkthrough. Start a new Chat, choose GPT-6, and select Medium where available. Leave Pro reasoning unselected. The Chat menu and thinking controls say GPT-6 rather than Sol.
There is no special activation prompt. Intelligent UI can appear when it helps answer your request.
Copy and paste this request:
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.
Keep the inputs and results together, with clear labels. 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 labels and starting values before clicking anything. The initial total should be $700, the cost per attendee $35, and the amount left in the budget $200.
The calculation is $200 + 20 × ($18 + $7) = $700. We independently calculated these figures and confirmed them in the live calculator.
Our first calculator worked but stacked the inputs vertically. To keep the inputs and results readable together, we sent this layout-only follow-up and repeated the checks:
Keep this calculator directly in the conversation, but make its layout compact and easy to read on a laptop: arrange the five inputs in two columns and the three headline results in one row, with a short visible formula and a reset control. Preserve every starting value, calculation, validation rule, and the zero-attendee behavior. Do not create a separate webpage or code preview.
When the answer is only text
Ask once:
Show editable inputs and a working calculator inside this conversation, rather than only describing the calculations. Keep the same values and rules. Tell me if this interface is unavailable here.If it still returns text, use the access checks below. Do not count a static table, generated image, downloadable webpage, or separate code preview as a successful run of this exercise. The specific layout you receive may differ. We did not need this fallback in our test.
Use the calculator’s controls to change attendees from 20 to 30. Use its calculate button if it has one; our version updated immediately.
With the room fee, per-person costs, and budget unchanged, expect:
| Result | 20 attendees | 30 attendees |
|---|---|---|
| Room fee | $200.00 | $200.00 |
| Food and materials | $500.00 | $750.00 |
| Total cost | $700.00 | $950.00 |
| Cost per attendee | $35.00 | $31.67 |
| Budget remaining | $200.00 | −$50.00 |
The per-person cost falls because more people share the room fee. The total still rises above the budget. Check both results before deciding that a larger group costs less.
Look for an explicit $50 over budget warning. A negative number should not appear under a label such as “money available” without explanation.
Check that the calculator preserved the other four inputs. Then restore the starting values and confirm that the $700 total returns. Both checks passed in our run.
Keep the original costs and no buffer. Try these values:
| Test | Expected result |
|---|---|
| Zero attendees | Total cost stays $200. Cost per attendee reads Not applicable. The budget has $700 remaining. |
| 28 attendees | Total cost is $900, exactly the budget. |
| 29 attendees | Total cost is $925, which is $25 over budget. |
| A negative or fractional attendee count | The control rejects it or asks for a whole, nonnegative number. |
| A blank input or negative dollar amount | The calculator shows an input error instead of a usable result. |
These cases passed in our test. Invalid inputs displayed a message and replaced the calculated results with dashes until the error was fixed.
The room fee stays at $200 even when attendance is zero because that is the rule in our example. Real cancellation terms could differ.
If a result fails a check, use a focused request such as:
At 29 attendees, the room fee is $200 and the per-person costs total $25. The correct total is $925. Explain why this calculator differs, then fix the calculation without changing the inputs. Recheck 20, 28, 29, and zero attendees.For a control that accepts a fraction, ask it to reject the value rather than silently rounding the headcount:
Reject negative, fractional, or blank attendee counts with a clear input error. Do not silently round the headcount or show a valid-looking total for an invalid value. Preserve the original costs, budget, and zero-attendee rule, then recheck valid and invalid inputs.We did not need these repair prompts. A neat result with the wrong assumptions still needs correction.
Return to 20 attendees and send:
Add an editable budget buffer percentage, starting at 10%. Calculate the buffer as 10% of the room, food, and materials subtotal. Add it once to get the final cost.
Show the subtotal, buffer amount, final cost, cost per attendee, and budget remaining. Keep the $900 budget, the original cost inputs, the zero-attendee rule, and input validation. The restore control should return to 20 attendees, the original costs, and a 10% buffer.At 20 attendees, the result should be:
| Item | Expected result |
|---|---|
| Original subtotal | $700.00 |
| 10% buffer | $70.00 |
| Final cost | $770.00 |
| Cost per attendee | $38.50 |
| Budget remaining | $130.00 |
Change attendance to 30 again. The subtotal should become $950, the buffer $95, and the final cost $1,045. That is $145 over budget, with a cost per attendee of $34.83.
This second check tells you whether the revision still responds to inputs. Inspect every result that depends on the new rule; a correct headline total can sit beside a stale per-person figure.
We confirmed both sets of results. At zero attendees, the revised calculator retained the $200 room fee, added its $20 buffer, showed Not applicable per attendee, and left $680 in the budget. Invalid-input checks still worked, and Reset restored the original costs, 20 attendees, and a 10% buffer.
Before using the output in a real plan, ask for a short text record. Copy the checked summary into your planning document:
Give me a short plain-text record of the restored calculator: 20 attendees, $200 room fee, $18 food and $7 materials per attendee, $900 budget, and 10% buffer. Include the inputs, formula, subtotal, buffer, final cost, cost per attendee, budget remaining, and assumptions. Keep these fictional figures separate from real bookings, payments, and messages.Our text record matched the checked inputs and results. Leave bookings, payments, and messages outside this exercise.
Keep going
Pro adds the full on-demand course library, certifications, group office hours, and eligible member perks.
$29/mo, cancel anytime
Intelligent UI has no separate quota. Your existing model, tool, and plan limits still apply. The practice task needs no plugin.
No. ChatGPT can still choose a plain-text response. Ask for controls when changing inputs or exploring details would help. OpenAI’s announcement describes how it chooses a useful format.
GPT-6 can start replying while it continues thinking or using tools, then add more detail. Read the completed result before relying on its conclusion, as described in the October 7 release notes.
On the web, open Settings and turn off Layout and visuals. This visual preference reduces visuals but may leave some; it is not a complete text-only switch. We verified the documentation rather than changing this setting during our calculator test.
For this exercise, you change a tool inside the conversation. Our Astra browser-use guide covers checking outside websites in ChatGPT Work. Pick that workflow when the task needs live pages rather than the fictional inputs used here.