ChatGPT Gets GPT-6 and an Interface That Builds Itself

OpenAI has started rolling GPT-6 out to everyone who uses ChatGPT, and it has given the chatbot a new trick on the way in. The feature is called Intelligent UI, and it lets an answer assemble itself out of more than text. A reply can now arrive with tappable buttons, a chart, a short form, a calculator, a map, even a small game, chosen by the model to suit the question. When plain prose does the job, you still get plain prose.

The release reached the paid tiers first. OpenAI's announcement, GPT-6 and Intelligent UI for everyone, says Plus, Pro, Business and Enterprise users began getting it in the Chat tab on October 7th, with the Free and Go tiers following from October 8th. Which model you talk to depends on what you pay for. Paid tiers run on GPT-6 Sol, while Free and Go run on GPT-6 Luna, the cheaper, faster sibling the two models were introduced as last month.

What actually changed

The interface is the headline, but the quieter change is how the model answers. GPT-6 can begin replying while it is still thinking or still using a tool, then keep adding to the answer without a fresh prompt. In practice that means you can start reading before the full response has landed. OpenAI pitches the whole package at the scale it now operates on, which it puts at more than 1.2 billion people using ChatGPT each week.

A few things are deliberately left alone. The models behind ChatGPT Work and Codex are not changing with this release. The Pro reasoning option still runs on GPT-6 Astra and does not support Intelligent UI, which makes sense: a maths-heavy deep-reasoning request is not the place for a bill-splitter widget. Intelligent UI is available across the reasoning range otherwise, from the quick Instant setting up to the slower, more deliberate modes.

A familiar idea, wired into the model

The concept of a chatbot generating a small interactive panel rather than a wall of text is not brand new. What is new is making the model itself decide when to reach for one, and shipping it to over a billion people at once. That is a bet about how most people want to use these systems: less like a command line, more like a tool that hands you the right small tool back.

It also fits the direction OpenAI set out at its developer day, when it leaned hard into agents that act before you ask and the cheaper, faster Sol line. The through-line is an assistant that does more of the assembling for you. Whether that feels like help or like a loss of control probably depends on the task, and on how often the model guesses the format wrong. The honest answer on that will come from a billion people trying it over the next few weeks, not from a launch post.