On October 7, OpenAI pushed Intelligent UI to everyone
On October 7, OpenAI began rolling out GPT-6 together with Intelligent UI to all ChatGPT users. Paid tiers (Plus, Pro, Business, Enterprise) got it the same day, and Free and Go tiers followed the next day. This is not a cosmetic refresh. It lets the model compose answers with interactive pieces inside the response: charts, buttons, forms, even small on-the-fly tools.
Not a reskin: interface generation moved into the model
Before, you asked a data question and got prose with a few bullets. Now the model decides a chart would be clearer and draws one, or decides a step needs a tap and renders a button. OpenAI frames this as interleaving thinking with answering: the model emits interactive fragments as it reasons, instead of holding everything back for one block of plain text at the end.
Two engines power it. Paid users run GPT-6 Sol, free users run GPT-6 Luna. The chat box that draws its own buttons is, underneath, two different model tiers doing the work, and the experience gap lives in the tier.
How it is priced: the feature is free, the cost is on the API side
For ordinary users, Intelligent UI is part of GPT-6 and carries no separate charge. The real cost sits on the API side: if your product calls this interactive layer, you pay Sol or Luna token rates, and while the components themselves are not billed separately, the inference tokens spent generating them are. For product teams, flash costs real money, so do not default to stuffing every reply with widgets.
Three colds showers it deserves
First, the components are model-generated and can be wrong. A chart the model draws on the spot may invent its axes, units, or data sources. People trust clickable things more, which amplifies the hallucination risk. For anything that matters, do not trust the picture; open the underlying data first.
Second, accessibility and export are open questions. Interactive blocks are unfriendly to screen readers and to anyone who needs to copy, archive, or drop content into a PDF. If your audience includes such users, load-test before shipping, or your goodwill becomes a wall.
Third, this is a moat, not an open standard. Intelligent UI locks the answer into ChatGPT's own rendering layer. The same content elsewhere may be plain text. Teams wanting cross-platform experiences should weigh the lock-in before handing over the keys.
An underrated signal: the answer is becoming the product
When the model draws charts and forms inside the reply, the line between chat and software blurs. A user who wants a figure used to get prose and then opened a spreadsheet elsewhere; now the spreadsheet renders in the conversation and is interactive. The answer itself is turning into a lightweight product, and the model looks more like an operating system than a search engine. For sites that rely on users clicking a link to leave, this is a structural shift.
For content-site owners, the traffic logic tightens another notch
Sites like ainspiro already see AI assistants driving direct traffic, a large share of that Direct bucket in analytics is really AI clicks. Intelligent UI only accelerates the trend: a user satisfied by an interactive card in ChatGPT may never open your source page. The response is not resistance but making structured, accurately citable facts solid, because being quoted by the model in conversation is more durable in the AI era than a visit to your page.
What this means for you
If you build content or tools, the signal is that plain-text answers are turning into interface-bearing answers, and your competitors may ship theirs next month. But do not chase the trend blindly. Decide whether your user genuinely needs to tap, or merely needs to see clearly. Only the first case earns interactivity; the second is fine with a static image. When unsure, start static: lowest cost, least rework.