GPT-6 lands in ChatGPT, building an interface for each answer

OpenAI started rolling GPT-6 out to ChatGPT on October 7, and the feature it leads with is not a model tier. It is a new kind of answer. With what the company calls Intelligent UI, the model composes its response out of text, graphics, charts, buttons and forms, picking the layout for each question. The rollout began on Plus, Pro, Business and Enterprise, and extends to Free and Go the next day.

This follows GPT-6 Astra and the GPT-6.1 Sol models that OpenAI shipped to paid customers and developers last month. This release targets everyone: OpenAI puts ChatGPT's weekly usage above 1.2 billion people.

What Intelligent UI actually is

GPT-6 was trained to assemble responses from text, visuals and interactive elements, and to decide how the pieces fit based on the question. In the company's own examples, a question about how a 7-speed bicycle works comes back as a tappable map of its systems, and a dinner plan comes back as a menu with an interactive control that rescales the quantities when the guest count changes. The announcement also shows small tools you can ask for inside the conversation: a savings calculator, a bill splitter, a game.

Under the hood, OpenAI built a library of native, streamable components plus a compiler that processes the interface as the model generates it. The UI therefore appears progressively instead of waiting for the full response. The training loop includes evaluating the interfaces the model builds for clarity, usefulness and completeness. OpenAI itself says the design judgment still has work ahead, and that plain text remains the format when plain text is the answer.

Answers that start before the thinking ends

The second thread is speed. Reasoning models improved the hardest problems, but users had to wait for the whole thinking pass before the first word arrived. GPT-6 interleaves thinking with answering: it starts composing a response from what it already knows and keeps refining as the reasoning continues, and the company says the model learned to weigh how long the user has been waiting.

The numbers come from OpenAI's internal evaluations, so read them with that context:

• On a set of everyday agentic tasks, GPT-6 Extra High begins answering in the same time as GPT-5.6 Medium, and scores better overall than GPT-5.6 Extra High.
• On questions that require web search, GPT-6 Instant begins answering 44% sooner than GPT-5.6 Instant, on average.
• OpenAI also claims GPT-6 makes better calls about when to search and more reliably finds the information that backs its answers.

None of these are third-party benchmarks. They are the company's own measurements, published in the announcement.

What changes for your setup

The practical part is availability. From October 7, GPT-6 with Intelligent UI ships in the Chat tab to Plus, Pro, Business and Enterprise users, and Free and Go join on October 8. Enterprise access still depends on workplace admin settings. Paid tiers run on GPT-6 Sol and the free tiers on GPT-6 Luna, both tuned for everyday conversation. The models behind Work and Codex do not change in this release, so API and coding workflows keep what they already have.

Safety notes

OpenAI says GPT-6 builds on several of Astra's safety advances. In its adversarial testing, the model resisted jailbreak attempts more strongly, especially attacks that adapt across multiple turns, and its alignment evaluations show it communicates its own limits better than GPT-5.6 Sol. The full results are in the system card published alongside the post.

If you are on a paid tier, the test is simple: ask for an answer that needs a comparison, a plan or a calculator, and check whether the interactive result is actually useful, or whether you end up fixing more than you would have typed.