OpenAI began rolling out Intelligent UI on 7 October: a calculator, buttons and diagrams instead of a wall of text, when the model judges it fits. Paid plans get GPT-6 Sol, and Free and Go, per OpenAI's plan, follow from the next day with GPT-6 Luna.
- 7 October 2026: OpenAI is rolling out a new GPT-6 model with Intelligent UI in ChatGPT. An answer can combine text, images, buttons, forms, charts and small tools, assembled on the spot.
- Access per the announcement: Plus, Pro, Business and Enterprise from 7 October with GPT-6 Sol; Free and Go, per OpenAI's plan, from the next day with GPT-6 Luna. Work and Codex are not changing with this release.
- According to OpenAI, ChatGPT can now start answering while it is still thinking: GPT-6 Instant starts on average 44% sooner than GPT-5.6 Instant on questions that need search. The company itself admits the model's design judgment still needs work.
Ask ChatGPT for a calculator for your savings and, instead of a formula in text, you may get a working calculator inside the chat itself. That is Intelligent UI, and it comes with GPT-6 in ChatGPT. The model decides what the answer looks like: the roast timing sits next to the recipe, the stops of a road trip appear on a map, the explanation is an interactive diagram where you change an input and see what happens. The examples are from OpenAI's announcement. When plain text is the most useful thing, you get text.
The free plan gets Luna, not Sol. Paid plans have Sol. OpenAI calls both tuned for everyday conversation, but how close they are in quality, the page does not say. The interface is for everyone, and the model depends on the plan.
On stage, the most valuable button is the one that sits in the same place every night. Here the model will choose where it goes. OpenAI does not hide it: the library of ready components is its answer, and the announcement carries an honest admission that the model's design judgment still needs work. The examples on the page were chosen by OpenAI. How often the interface will help and how often it will get in the way, it does not say.
The second half of the news is quiet. ChatGPT can now start answering while it is still thinking. The 44% earlier start on questions that need search, though, is OpenAI's own figure, with no method described, and the page has no outside check.
The test is the question you ask every day, not the roast from the demo. That is where it will show whether the interface helps or just takes up space.