Ask ChatGPT to split a dinner bill and, until recently, you got a paragraph of arithmetic you had to recheck yourself. Since October 7, 2026, you might get something else entirely: a working bill splitter with sliders you can drag, built right inside the answer. OpenAI calls this Intelligent UI, and it shipped with GPT-6, the newest model family behind ChatGPT. Instead of answering everything in text, GPT-6 decides — question by question — whether your answer should be a paragraph, a chart, a side-by-side comparison, an interactive diagram, or a small tool you can actually use.
This is the biggest visible change to how ChatGPT feels since the chatbot went mainstream, and it is already on every tier, including free. But it is not a magic switch, it does not work in every mode, and the model still gets things wrong — sometimes inside a very convincing-looking calculator. Here is what Intelligent UI actually does, who gets it, where it falls short, and the prompts that make it useful instead of gimmicky.
What Intelligent UI is, in plain terms
Intelligent UI is GPT-6’s ability to compose a reply from text, visuals, and interactive elements together, choosing the format based on your question. OpenAI’s October 7 announcement puts it this way: a comparison might work best side by side, an explanation might be easier through an interactive diagram, and when a simple text answer is the most useful response, ChatGPT still gives you that.
The examples OpenAI itself showed cover a wide range:
- A savings calculator or a dinner bill splitter built on the spot for your exact numbers
- A recipe with the cooking timeline shown next to the steps
- Road trip stops placed on a map
- An interactive diagram of a seven-speed bicycle, where tapping a part explains what it does
- Interactive explanations of concepts like the central limit theorem and the Monty Hall problem
- A small retro game you can play inside the conversation
Under the hood, OpenAI says the feature rests on two pieces: a library of native, streamable interface components and a compiler that assembles the interface while the model is still generating. That is why a chart or tool can start appearing before the full answer finishes, instead of waiting for a finished block of text. GPT-6 also introduces what OpenAI calls interleaved thinking — it can begin answering while it continues to think or use tools, then add findings as they arrive. According to OpenAI’s internal evaluations, GPT-6 Instant begins responding to questions that need a web search an average of 44% sooner than GPT-5.6 Instant did.
The important mental shift: you do not ask for a tool. You ask a question, and the model picks the format. That is both the appeal and the first frustration, which we will get to below. Contemporary coverage of the launch — including PYMNTS’ report on the October 7 announcement and ghacks’ rollout breakdown — confirms the same two-wave schedule and the same Pro-reasoning exception described here.
Who gets GPT-6 and Intelligent UI, and when
The rollout happened in two waves. Paid tiers — Plus, Pro, Business, and Enterprise — got GPT-6 on October 7, 2026. Free and Go users followed the next day, October 8. There was no price change; GPT-6 replaces the existing “Latest” option in the model picker.
Which model you actually get depends on your plan:
| Plan | Model in the Chat tab | Intelligent UI |
|---|---|---|
| Plus, Pro, Business, Enterprise | GPT-6 Sol | Yes |
| Free and Go | GPT-6 Luna | Yes |
| Pro reasoning option | GPT-6 Astra | No |
That last row is the catch most coverage buries. If you switch ChatGPT into its Pro reasoning mode, you are running GPT-6 Astra, and OpenAI’s own release notes are explicit: the Pro reasoning option “doesn’t support Intelligent UI.” Intelligent UI is available from Instant through Extra High reasoning, but not in the Pro reasoning setting. So the most expensive reasoning mode is the one place the new interface does not show up.
Two more scope notes. First, the rollout applies to the regular Chat tab only. The models behind ChatGPT Work and Codex are unchanged by this release. Second, the rollout is staged, so if your account still shows the older model, that is normal — it may simply not have reached you yet. OpenAI puts ChatGPT’s weekly audience at more than 1.2 billion people, which is a lot of accounts to flip over in a week.
If you have been following the GPT-6 story on this site, the free-tier side of the rollout connects directly to our earlier piece on GPT-6’s free-tier rollout, and the Sol/Luna/Codex split is covered in our GPT-6 model guide.
The tested prompt set: where interactive answers actually help
OpenAI’s demos are polished, but the real test is boring everyday math — the stuff where a wrong number costs you money or an argument with friends. An independent YouTube tester ran GPT-6 Luna on the free tier through three real tasks, and the results are a useful reality check:
- Bill splitter: an $86 dinner bill with an 18% tip split four ways came out at $25.37 each — correct. Five friends at 20% came out at $20.64 each — also correct.
- Phone comparison: a side-by-side iPhone 17 vs Pixel 10 comparison pulled in web results and photos and built a clean table with sources.
- Mortgage calculator: a $300,000 loan at 6.5% over 30 years came out at $1,896.20 a month with $382,633 in total interest — exactly right — and showed that paying $200 extra a month cuts the loan to 23 years and 1 month and saves about $103,449 in interest. The tester checked every number independently.
That is the strongest version of Intelligent UI: a small tool built for your exact numbers, where you can change an input and watch the answer move, instead of re-asking the whole question. Here are the prompts that play to that strength, adapted from the tested set:
- Split a bill: “Split an $86 dinner bill between 4 friends with an 18% tip. Make it interactive so I can change the numbers.”
- Compare two products: “Compare iPhone 17 vs Pixel 10 side by side: battery, camera, price, and who each is best for.”
- Test a loan: “Show me a mortgage calculator for a $300,000 loan at 6.5% over 30 years, and let me change the extra monthly payment to see the interest saved.”
- Learn something mechanical: “Explain how a seven-speed bicycle works with an interactive diagram I can tap through.”
- Plan a trip: “Map a road trip from Denver to Austin with stops, and show the timeline next to the route.”
- Grow savings: “Build a savings calculator for $100 a month and let me adjust the monthly amount and the target.”
Notice the pattern in the ones that work best: a concrete set of numbers, plus an instruction to make the result adjustable. The phrase “make it interactive” turns out to matter — the tester found that adding those words reliably triggered the interactive format, while a plain question sometimes produced plain text. When the format is the model’s decision, a nudge in the prompt is your only lever.
When plain text still wins
Interactive is not automatically better, and GPT-6 knows it — sometimes. A one-fact question (“What time does the pharmacy close?”) does not need a dashboard. A nuanced opinion or a sensitive explanation usually reads better as prose. And there are places where an interactive element actively hurts: a chart with a wrong axis, a calculator with a wrong formula, or a diagram that looks authoritative while quietly skipping a step.
OpenAI is unusually candid about this. Its announcement says there is “still work ahead to improve the model’s design judgment and expand what it can create.” In practice, that means the model sometimes picks a format that adds flash without adding understanding, or builds a tool whose inputs do not cover the edge case you actually care about (tax on the bill, fees on the loan, a variable rate instead of a fixed one).
My rule of thumb after looking at the tested cases:
| Situation | Best format | Why |
|---|---|---|
| Numbers you will change and recheck | Interactive calculator | You test scenarios without re-asking |
| Two or more options with trade-offs | Side-by-side comparison | Differences are visible at a glance |
| A process with moving parts | Interactive diagram | Tapping beats re-reading a wall of text |
| One fact, one definition, one opinion | Plain text | Anything else is decoration |
| Money, health, or legal decisions | Text plus a tool you verify | The tool helps you explore; it does not decide for you |
The last row matters most. A calculator with the wrong formula is still wrong, no matter how good it looks. The mortgage example above checked out exactly, but that was one tester’s verified case, not a guarantee. For anything where the number has consequences, run the result past a second source — your bank’s calculator, a spreadsheet, or the actual bill — before you act on it.
GPT-5.6 vs GPT-6: what actually changed
It is easy to treat GPT-6 as “GPT-5.6 plus widgets.” The difference is more specific. GPT-5.6 was built around reasoning, tool use, coding, and long tasks, and it answered primarily in conversational text. GPT-6 keeps the reasoning but adds a second job: deciding what form the answer should take. It also answers while it is still thinking, instead of reasoning first and responding after, which is why web-search answers start arriving sooner.
GPT-5.6 also remains selectable in ChatGPT, so if you prefer the older behavior, you are not forced onto the new interface. That matters if you use ChatGPT for long writing or coding sessions where an unexpected chart would just get in the way.
The catches nobody puts in the headline
Four things worth knowing before you build a workflow around Intelligent UI:
1. Free-tier quirks. The same tester who verified the calculators found that free chats can show ads, that asking for a “diagram” sometimes triggered image creation (which needs a free login), and that a login pop-up appears after a few chats — closable, but persistent. In Settings, you can turn off personalized ads and turn off using your chats to improve the model. If you use the free tier for anything personal, check those two settings first.
2. No toggle. There is no “Intelligent UI on/off” switch. You can nudge it with words like “make it interactive,” and you can ask for plain text when you want it (“just give me the answer in text”), but you cannot pin the behavior. If your workflow depends on always getting a tool, that unpredictability is a real limitation.
3. Pro reasoning excludes it. As noted above, the Pro reasoning option runs GPT-6 Astra and does not support Intelligent UI at all. If you pay for Pro mainly for the hardest reasoning tasks, you will not see the new interface in the mode you use most. OpenAI has not said when or whether Astra gets it.
4. Design judgment is still early. OpenAI’s own admission is the honest summary: the model is still learning when a visual helps and when it distracts. Expect some answers that are prettier than they are useful, especially on unusual or multi-part questions.
How to get the most out of it this week
If you want to try Intelligent UI properly, here is the short version:
- Open a new chat in the regular Chat tab (not Work, not Codex).
- Check the model picker: paid plans should show GPT-6 Sol, free and Go should show GPT-6 Luna. If you see the Pro reasoning option selected, switch it off — that mode has no Intelligent UI.
- Start with a numbers question you already know the answer to, so you can spot a wrong formula immediately. The bill splitter above is a good first test.
- Add “make it interactive” to the prompt when you want a tool, and “just text” when you do not.
- Change an input and watch the result update. If it does not update, or the math looks off, ask for the formula in text and check it.
- For money decisions, verify against a second calculator before acting.
Intelligent UI is not a reason to switch plans on its own — free users get the same capability as paid ones, on the lighter Luna model — but it is a reason to change how you ask. The chatbot that used to hand you a paragraph can now hand you a tool. The skill, as always, is knowing when the tool is worth trusting and when the paragraph was fine all along.
What this means for small businesses and creators
For a small business owner or a creator, the practical value of Intelligent UI is not the novelty of a chart — it is the five-minute task that stops needing a separate app. A freelancer can price a project by asking for a rate calculator with tax and expenses baked in, then slide the hourly rate until the monthly target makes sense. A seller can compare two shipping options side by side without opening a spreadsheet. A content creator can turn a “which camera should I buy” question into a comparison table with prices, then ask for the same table again when a new model drops.
The flip side is that anything you publish or quote from an AI-built tool needs the same check you would give a paragraph. If a calculator tells you a price, a margin, or a tax figure, verify it before it goes into a quote, a listing, or a client email. The interactive format makes wrong numbers feel more finished than a paragraph does, which is exactly why the verification step matters more, not less.
One more practical note: because the free tier includes Intelligent UI, there is no reason to pay for a plan just to get it. Pay for Plus or Pro when you need higher limits, more image generation, or heavier reasoning — not for the interface. And if you run a team, remember the rollout is staged and the Work tab is unchanged, so do not promise a client or a colleague an interactive answer until you have seen it on your own account.
Frequently Asked Questions
What is ChatGPT Intelligent UI?
Intelligent UI is GPT-6’s ability to answer with interactive interfaces — charts, tappable buttons, forms, diagrams, and small working tools like calculators and bill splitters — instead of text alone. The model chooses the format based on your question, and still uses plain text when that fits best.
When did GPT-6 with Intelligent UI launch?
OpenAI began rolling out GPT-6 on October 7, 2026 for Plus, Pro, Business, and Enterprise users, and extended it to Free and Go users on October 8, 2026. There was no price change.
Does Intelligent UI work on the free plan?
Yes. Free and Go users get GPT-6 Luna, which includes Intelligent UI. Free chats may show ads, and some diagram requests can trigger image creation that requires a free login.
Why don’t I see interactive answers in Pro reasoning mode?
The Pro reasoning option runs GPT-6 Astra, which does not support Intelligent UI. Switch out of Pro reasoning into the standard GPT-6 Sol chat to get interactive answers.
Can I turn Intelligent UI off?
There is no on/off toggle. You can ask for plain text (“just give me the answer in text”) or nudge the interactive format with “make it interactive,” but the model ultimately decides the format.
Should I trust numbers from an AI-built calculator?
Use it to explore, then verify. Check money, health, and legal numbers against a second source — your bank’s calculator, a spreadsheet, or the original bill — before acting on them.