What the source published

OpenAI announced GPT-6 rolling out to everyone, introducing Intelligent UI: ChatGPT can compose responses from text, visuals and interactive elements, including graphics, tappable buttons, forms, charts and interactive experiences you can use directly in the conversation. The format follows the question — a comparison works best side by side, an explanation is easier through an interactive diagram, and when plain text is the most useful response, you get plain text.

The everyday examples are concrete: for a Sunday roast, the timing can sit alongside the recipe so the plan is easy to follow; the stops on a road trip can appear on a map, with notes on what is worth a detour. For learning, visual and interactive explanations let you change an input to see what happens, or work through a problem step by step. And you can ask it to create a tool for the task at hand, such as a calculator to explore how savings could grow, a bill splitter for dinner with friends, or a game you can play in the conversation.

The rollout covers the Chat experience: Plus, Pro, Business and Enterprise run on GPT-6 Sol, while Free and Go run on GPT-6 Luna, with paid tiers first and Free and Go following a day later; Enterprise availability depends on admin settings. The models powering Work and Codex are not changing in this release.

Sources: [1] [3]

Try it: four prompts that unlock interactive answers

  1. For comparisons, ask for side by side explicitly: say you want the two options compared with the same dimensions in the same rows, and it will lay them out as a table or paired cards instead of two paragraphs to cross-read.
  2. For learning, ask for something you can poke: request a diagram with an input you can change and results that update as you do, so you build intuition by experimenting rather than rereading.
  3. For small jobs, have it build the tool: savings growth, bill splitting, unit conversion — describe the inputs and the output you want and let it generate a tiny calculator on the spot.
  4. For plans, ask for time and steps together: a meal or a trip works better when the schedule and the content share one view, instead of you flipping between a list and a clock.
  5. For hard questions, do not interrupt early: this generation answers while it thinks, and the first partial answer often carries the key point — wait for the complete version, then correct the parts you dislike.

Sources: [1]

The full picture: how the interface streams into existence

Under the hood, OpenAI built a library of native, streamable components and a compiler that processes the interface as the model generates it. The library gives each response a familiar design foundation while leaving the model room to decide how the pieces fit together, and the compiler lets the interface appear progressively rather than waiting for the entire response to finish. Training was expanded too: interfaces are evaluated for clarity, usefulness and completeness, so the model learned how to organize information, when to use interactivity, and when a simple text response is enough.

The speed change comes from interleaving thinking with answering. Instead of finishing all reasoning before speaking, ChatGPT progressively composes an answer from what it knows so far, with each partial response adding useful information. In OpenAI's internal evaluation, GPT-6 Extra High begins answering in about the same time as GPT-5.6 Medium while scoring better than GPT-5.6 Extra High; on questions needing web search, GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant, decides better when a search is needed, and finds information that supports the answer more reliably. On difficult problems it addressed the key aspect of the question more often, and it is clearer about what it cannot do or does not know.

Sources: [1] [3]

How to read this

The real shift is that an answer becomes a temporary interface instead of a paragraph. Your prompting should follow: it is no longer enough to state the question clearly; say what you intend to do with the answer — compare, simulate, execute — and the format follows. Left unspecified, the model picks a format for you, and its default may not be the one that serves you best.

Keep one habit, though: an interactive surface feels more finished, which is not the same as more correct. For numbers you will act on — money, dates, quantities — the slicker presentation changes nothing about verification. Treat this as a big step in form, and keep your checking routine exactly where it was.

Sources: [1]

Key takeaways

  • Intelligent UI composes answers from text, graphics, buttons, forms and charts, choosing the mix per question — and you can request a format explicitly.
  • The three most useful moves: side-by-side comparisons, diagrams with inputs you can change, and throwaway tools like calculators and bill splitters generated on request.
  • The rollout touches Chat only: paid tiers get GPT-6 Sol, Free and Go get Luna, and the Work and Codex models are unchanged.
  • Answer-while-thinking cuts the wait; on search questions answers start 44% sooner on average, and hard questions more often hit the key point.
  • A polished interface is not a verified one: still check the underlying numbers before acting on money or dates.

Sources: [1]