Ask ChatGPT Intelligent UI for Charts On Cue

OpenAI began rolling out Intelligent UI inside ChatGPT on October 7, 2026, alongside GPT-6, and its own demo centers on a seven-speed bicycle diagram with tappable parts, according to Emma Roth’s report for The Verge. The update lets the model answer with interactive visuals in the same thread as the prose.

When a reply should be a chart, a form, or a small tool, OpenAI says it trained GPT-6 when to generate interactive visuals over text, as well as how to format them in its response, as Roth relays from the company’s launch blog post.

Prompt diagrams buttons and inline tools

Open Intelligent UI the same way you open any ChatGPT turn: go to chatgpt.com (or the mobile app), sign in, and make sure a GPT-6 model is selected for the chat. Plus, Pro, Business, and Enterprise accounts started receiving the mid-range GPT-6 Sol model with the feature on October 7. Go and free accounts were scheduled to receive the more efficient GPT-6 Luna model on the following Thursday, still per The Verge’s rollout note. If your picker still shows an older model, switch to Sol or Luna before you chase a missing widget.

There is no dedicated Intelligent UI switch in the coverage that shipped with the launch. OpenAI says it trained GPT-6 when to prefer interactive visuals over plain text, and how to format those surfaces inside the reply. Your lever is the prompt. Ask for a design, a layout, or a manipulable object when a paragraph would bury the structure. The bicycle demo is the clean pattern: ask how a seven-speed bicycle is put together, and the model can return a diagram plus buttons that highlight parts. Ask how to play Mahjong, and it can generate tile visuals sorted into scrollable categories. Ask for a retirement savings calculator, a bill splitter, or a small retro game, and OpenAI’s examples show the tool generated in-line with the answer rather than as a separate download.

Use that pattern on ordinary work without inventing a special syntax. Name the artifact you want in the first sentence. “Build an interactive bill splitter for four people with tip and tax fields” is clearer than “help me split a bill,” because it tells the model the surface is part of the deliverable. “Draw a labeled diagram of X and let me tap each part” does the same job for spatial explanations. If you only want prose, say so up front: “Answer in plain text only, no charts or forms.” The model still decides, but an explicit constraint is the closest thing to an off switch the public write-ups describe.

When a visual does appear, treat it as part of the conversation, not a one-shot image dump. Tap or click the controls the reply exposes, then ask a follow-up that refers to what you changed (“recalculate with a 20% tip,” “highlight only the drivetrain”). Intelligent UI is useful when the next question depends on a value or a region you just adjusted. Steer toward OpenAI’s own demos when you want that loop: a bill splitter you can retune, a bicycle diagram with parts you can highlight, or Mahjong tiles you can scroll by category. Keep the ask textual when you want a long prose answer instead of an in-line tool.

OpenAI also says GPT-6 can provide quicker, partial answers while it gathers information, per the same Verge report. That claim sits with the GPT-6 ChatGPT release alongside better web search, not as a published rule about how Intelligent UI panels load. If a panel stays blank, the launch coverage does not document a recovery path. Untested editorial advice only: regenerate once, ask for a single widget, or request a plain text explanation with a simple Markdown table.

Weigh OpenAI’s trained-when claim cold

OpenAI’s headline claim, relayed in that same Verge piece via the company’s launch blog post, is that GPT-6 was trained both on when interactive visuals beat text and on how to format them in the response. The supporting exhibit set is OpenAI’s own: the seven-speed bicycle, the Mahjong tile browser, the retirement calculator, the bill splitter, the retro game. No third-party task mix, no public win-rate, and no disclosed comparison against GPT-5-era ChatGPT sessions appears in the launch coverage. The claim buys you a mental model (the model will sometimes choose a UI without you naming a product feature), not a measured guarantee that your next prompt will surface one.

The training story leaves no settings checkbox for Intelligent UI, so “I asked for a chart and got an essay” is a model judgment you can only steer with clearer prompts.

OpenAI also ties the GPT-6 ChatGPT drop to stronger web search, quicker partial answers while information is still loading, and “stronger resistance to attempts to bypass its safety training.” In the Verge piece, those lines are GPT-6 bundle claims. A safer refusal policy does not prove a calculator will appear. Better search does not prove a diagram will be accurate. Hold the interactive-UI claim to the demos and the “trained when” sentence, and treat the rest as adjacent GPT-6 marketing until OpenAI publishes something you can re-run.

Match Sol Luna and plan access first

Access is plan-gated, not prompt-gated. As of the October 7 start described by The Verge, Intelligent UI rides with GPT-6 for ChatGPT Plus, Pro, Business, and Enterprise users first. Go and free tiers follow on the stated Thursday expansion, on Luna rather than Sol. If you are on a paid seat and still cannot raise a diagram after a direct ask, confirm the model picker shows GPT-6 Sol. If you are on free or Go before that Thursday window, Luna with Intelligent UI is simply not the build you are talking to yet.

Sol versus Luna is the other gate that matters for this tutorial. Higher-tier subscribers get Sol. Go and free users get Luna. OpenAI presents Luna as the more efficient option. The Verge does not publish a side-by-side of how often each tier emits interactive UI, or whether Sol is quicker to choose a form over a paragraph. For day-one use, that means you learn the prompt patterns on whichever GPT-6 your plan actually has, and you do not assume a free Luna session will mirror a Plus Sol demo screenshot from OpenAI’s marketing set.

Business and Enterprise users are on that same October 7 start with GPT-6 Sol and Intelligent UI, alongside Plus and Pro, according to The Verge. The report does not describe a separate Business or Enterprise product line for the feature, and it does not document an admin control that turns Intelligent UI off on its own. For those plans, the practical gate in the launch coverage is whether GPT-6 Sol is the model on the chat.

OpenAI has not published the share of GPT-6 ChatGPT turns that emit an Intelligent UI surface, so nobody outside the company can reproduce how often the feature fires on ordinary prompts.