OpenAI began rolling out GPT-6 in ChatGPT on 7 October with a feature it calls Intelligent UI, which lets the chatbot answer with buttons, forms, charts and small working tools instead of a block of text. According to OpenAI’s announcement post, the model picks the format that fits the question. A savings calculator, a dinner bill splitter or a map of road trip stops can appear inside the conversation itself.
The audience figure needs care. OpenAI says more than 1.2 billion people use ChatGPT each week, but that describes the user base, not the number who received the feature. The rollout is staged, and the tiers get different models. Plus, Pro, Business and Enterprise customers got GPT-6 Sol on 7 October. Free and Go users were due to start the following day, on a model called GPT-6 Luna. OpenAI describes both as built for ordinary chat and does not say how they differ. Enterprise access also depends on workplace admin settings.
The release is also narrower than the name suggests. The post says the update covers the Chat experience only, and the models behind Work and Codex stay as they were. GPT-6 itself reached paid customers last month, so what is new here is the version built for the mass-market chat product.
The engineering is more constrained than “the model designs an app” implies. OpenAI supplied a stock of pre-built interface pieces for the model to pick from, plus a compiler, a program that turns the model’s output into on-screen elements while the output is still being written. The model chooses and arranges ready-made pieces, and the screen fills in progressively instead of waiting for the full response. That keeps the look familiar across web and mobile and limits how badly a layout can go wrong. OpenAI concedes the weak spot: “There is still work ahead to improve the model’s design judgment and expand what it can create.”
The second change is speed. Earlier reasoning models made people wait until the thinking was done. GPT-6 starts replying while it keeps working, adding to the answer in several partial responses, and OpenAI says it trained the model to weigh how long the user has been waiting.
The numbers attached to this come from OpenAI’s own internal evaluations. In one test of common multistep jobs, OpenAI says GPT-6 at its Extra High setting started answering as quickly as GPT-5.6 at Medium, yet still outscored GPT-5.6 at Extra High. Where a question calls for a web lookup, GPT-6 Instant begins its reply 44 percent earlier than GPT-5.6 Instant does, averaged across OpenAI’s tests. A third claim, that GPT-6 more often addressed the key point of difficult questions than GPT-5.6, arrives with no figure at all. The post names no outside tester, task list or scoring method.
Safety claims follow the same pattern. OpenAI says GPT-6 held up better in adversarial testing, especially against “attacks that adapt across multiple turns,” and that its training now targets cyberattacks, biological threats and violence. The post publishes no pass or refusal rates and points readers to a system card for detail.
For anyone who ships a lightweight calculator, converter or explainer page, the competitor is now a free chat reply that builds one on request. The Free tier gets that ability on Luna, a model whose gap to Sol the post does not put a number on.
OpenAI, “GPT-6 and Intelligent UI for everyone,” company announcement post, 7 October 2026.
