By ChatGPT stopped being a text box this week. OpenAI shipped GPT-6 with Intelligent UI on October 7 — a system that replaces plain prose answers with dynamically generated interactive components. Calculators, sortable tables, trip planners, bill splitters, in-chat games: all rendered inline as the model generates them, live, with no page navigation required. Free and Go users got access today. Paying users got it yesterday.
How It Actually Works
Intelligent UI is not ChatGPT generating raw HTML. OpenAI built a library of pre-approved native components — buttons, forms, charts, timelines, calculators — and a compiler that renders them progressively as GPT-6 generates a response. The model selects and configures components from the catalog; the compiler turns that configuration into an interface. You see it appear piece by piece, the same way text streams today.
This architecture matters. By restricting the model to a pre-approved component catalog, OpenAI limits what the interface can do — and what it can be manipulated into doing. Compare this to Claude Artifacts, where the model generates raw HTML and CSS inside a sandboxed iframe, or the Vercel AI SDK’s generative UI, where developers hardcode which React component maps to which tool call. OpenAI’s bet is centralized control: the model decides layout and interaction type, but only from a known set of shapes. A detailed breakdown of all four approaches to generative UI — including their trade-offs and security risks — is worth reading if you are building in this space.
Two Models, One Feature
The rollout shipped two model tiers. GPT-6 Sol goes to Plus, Pro, Business, and Enterprise. GPT-6 Luna goes to Free and Go. Both models share the same 1 million token context window. Sol runs at $2 per million input tokens and $10 per million output tokens via the API. Luna runs at $0.10 input and $0.50 output. There is no separate charge for Intelligent UI access.
| Plan | Model | Rollout Date | API Input Price |
|---|---|---|---|
| Plus / Pro / Business / Enterprise | GPT-6 Sol | October 7, 2026 | $2 / 1M tokens |
| Free / Go | GPT-6 Luna | October 8, 2026 | $0.10 / 1M tokens |
Enterprise admins: Intelligent UI is off by default for your tier. Check your admin panel before users start asking why their ChatGPT looks different from everyone else’s.
What It Looks Like in Practice
The examples OpenAI demonstrated are mundane in the best way: a recipe that adjusts serving quantities when you change a number, a bill splitter that calculates per-person cost dynamically, a trip planner that renders an interactive map with stops, comparison tables you can sort. These are exactly the kinds of small utility tasks that currently require opening a separate app or Googling a calculator. That is the point. OpenAI is describing a future where a conversational interface handles “more of the small jobs” currently distributed across dozens of specialized tools.
Performance is up too. Web search queries get an initial response 44% sooner on average compared to GPT-5.6 Instant. GPT-6 Extra High matches GPT-5.6 Medium speed while scoring above GPT-5.6 Extra High on quality. The practical effect: faster answers and smarter decisions about when to search.
What Developers Need to Know
Here is the part most coverage buries: Intelligent UI is not in the API. If you build on GPT-6 Sol or Luna via the Responses or Chat Completions API, your users get the model’s intelligence — they do not get the interactive rendering layer. Work and Codex models are also unchanged. Intelligent UI is a ChatGPT product feature, not an API primitive.
That means if you want AI-generated interactive interfaces in your product today, you are still doing it the old way: Vercel AI SDK with your own component mappings, Claude Artifacts in a sandbox, or your own framework. OpenAI has not opened the component catalog to external developers yet. The full GPT-6 Sol and Luna pricing breakdown is useful if you are evaluating API cost for production workloads.
The Honest Caveat
OpenAI said it directly in the launch post: “work remains ahead to improve the model’s design judgment and expand what it can create.” That is unusually candid. Early analysis from the security research community backs it up. Studies of LLM-generated UI components found that more than half of AI-generated e-commerce interfaces contained deceptive patterns when business goals were present in the prompt. At 1.2 billion weekly ChatGPT users, even a 97% injection-resistance rate translates to a meaningful failure surface. State consistency is another rough edge — a calculator and a timeline in the same response can generate contradictory assumptions about the same data.
None of this kills the feature. It means the component catalog approach is the right architectural call for now — and that the rough edges are real, not theoretical.
The Bigger Shift
Text-only AI answers now have an expiration date. Intelligent UI is the first credible signal that the conversational interface model — one chat window that generates the tool you need, rather than opening a separate app — is becoming viable at scale. Whether users actually prefer it over dedicated tools will take months to understand. But the architecture is sound, the rollout is global, and the price is zero on top of existing plans. Read the official GPT-6 for Everyone announcement for OpenAI’s full framing.
For developers, the immediate question is not whether to adopt Intelligent UI — it is not available in the API yet. The question is whether the component catalog model, once exposed to external builders, changes how you think about what a product UI needs to be.
