OpenAI announced on October 7 that GPT-6 is becoming the default model for all ChatGPT users, alongside a new response interface called Intelligent UI. The company's official account put it this way:
"GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone."
The rollout comes in two stages. Plus, Pro, Business and Enterprise switched over the same day, running on GPT-6 Sol underneath; Free and Go tiers are receiving it starting the next day, on the cheaper GPT-6 Luna. The update only touches the Chat tab — the models used in Work and Codex stay unchanged.
Replies now grow their own controls
Intelligent UI is the visible part of this update. OpenAI says it trained GPT-6 to decide on its own how to lay out a given answer: plain text when that's enough, or charts, clickable buttons, forms, checklists and maps embedded right in the reply.
The examples OpenAI and reporters have demoed are everyday ones. Ask for a recipe and the reply can include a small widget that rescales ingredient amounts by number of servings; ask about saving money and it can build a compound-interest calculator on the spot; after dinner with friends, it can produce a bill-splitting tool; it can even generate a small playable game inside the conversation. When ChatGPT asks a follow-up question, it can also offer clickable options instead of plain text.
Technically, OpenAI describes this as a native, streamable component library paired with a compiler that builds the interface in step with the model's own generation. That means the reply doesn't have to wait for the full response to finish before rendering — GPT-6 can push out the first part of an answer while it's still thinking or calling tools. OpenAI's own figure: for questions that need a web search, GPT-6 starts answering on average 44% faster than GPT-5.6 Instant.
A two-week rollout
Lined up on a timeline, this GPT-6 generation has shipped unusually fast.
Sol and Luna reached the API on September 22, priced at half of their GPT-5.6 equivalents; at that point only Work, Pro and Business tiers in ChatGPT could select them, with Free and Go getting Luna on desktop only. On September 24, ChatGPT's voice mode gained the ability to switch between all three GPT-6 models. On September 30, GPT-6.1 Sol launched, with API pricing cut to a fifth of Astra's.
By October 7, GPT-6 moved from "optional" to "default," suddenly covering every free user at once. OpenAI has not disclosed any usage or cost figures on whether the inference cost of moving free users to Luna is sustainable.
Everyone is building "answers that lay themselves out"
OpenAI is actually a late mover here. Anthropic built Artifacts into Claude back in 2024, rendering code, documents and small apps in a separate pane beside the conversation; Google introduced Dynamic View in the Gemini app alongside Gemini 3 last November, also letting the model generate an interactive page based on the question.
The approach differs, though. Artifacts and Gemini's Dynamic View behave more like a separate canvas opened outside the conversation, usually triggered deliberately or in a specific mode; OpenAI's version embeds components directly into an ordinary reply, with the model deciding by default whether to use them. The former leans toward "building something," the latter toward "handing you a tool while answering a question."
Turning it on by default raises a new problem too: a broken component is harder to spot at a glance than a wrong sentence — a calculator that gets the interest rate wrong looks no different from one that gets it right. OpenAI hasn't explained how the calculation logic inside these components is verified, and there's no mention in public materials of whether users can turn Intelligent UI off in settings.
Sources: OpenAI's official account announcement, MacRumors, CocoLoop, Search Engine Journal; model-to-tier mapping, rollout order and the 44% speed figure checked against OpenAI's own announcement.