OpenAI is bringing GPT‑6 to everyone. The company says a new GPT‑6 model is rolling out to the more than 1.2 billion people who use ChatGPT each week, a month after the first GPT‑6 models went to paid customers.
The headline feature is called Intelligent UI. Instead of answering with plain text, ChatGPT can now compose responses using graphics, tappable buttons, forms, charts, and other interactive elements, choosing the format based on the question. A comparison might appear side‑by‑side; an explanation might arrive as an interactive diagram. When plain text is the most useful answer, OpenAI says you'll still get plain text.
The demo problem
OpenAI chose a Sunday lamb‑roast plan to showcase the feature. In the blog post the answer appears under a "GPT‑5.6 Instant" header, with a "GPT‑6 Instant" label later in the page, making it unclear which model generated which part. Either way, the interactive payoff does not survive in a static write‑up - the announcement relies on a visual experience that can't be fully captured in plain text.
The recipe itself is solid: a scaling table for guest counts, a cooking schedule, and suggested follow‑up actions you could tap. The content demonstrates how the model can blend text with structured data, but the lack of a live UI in the article means readers must trust the description without seeing the interaction.
Tools on demand
Further down, OpenAI notes that you can ask ChatGPT to build a tool in the moment - a calculator for watching savings grow, a bill‑splitter for dinner, or even a simple game you can play inside the conversation. This is the most interesting promise, but also the riskiest. A chatbot that occasionally hallucinates is one thing; a generated calculator that returns incorrect numbers while looking authoritative is another. The announcement does not provide statistics on how often these generated tools are accurate, which is worth monitoring as the feature reaches a massive audience.
How it works
OpenAI built a library of native, streamable components plus a compiler that renders the interface progressively as the model generates it, so users aren't left staring at a spinner waiting for the full response. The model selects pieces from the library and decides how they fit together; it does not write arbitrary code into the chat window.
This design choice mitigates security concerns. Shipping raw code to a billion‑person audience would be a major risk, so treating the UI as a templated system is a prudent approach.
One oddity worth noting: the copy of the blog post cuts off mid‑sentence at "We expanded our training met." The missing text likely contained additional details about the training process, but it isn't available in the current version.
What I'm watching
The real story is scale. Forms and buttons are how software gathers input, and ChatGPT can now place them in front of 1.2 billion weekly users. This is a large‑scale experiment in how much people will trust software generated on the fly. We'll need to see how reliable those interactive tools are and how OpenAI addresses any misuse.