OpenAI rolls out GPT-6 in ChatGPT with Intelligent UI: answers now come with buttons

OpenAI is rolling out GPT-6 to ChatGPT users worldwide today, paired with a feature called Intelligent UI that replaces plain-text answers with interactive ones. Paying customers get GPT-6 Sol and free users get GPT-6 Luna, according to The Decoder.

The headline is the interface. The numbers underneath it are thinner than the buttons.

What Intelligent UI does

Per The Decoder, the model now presents answers as interactive interfaces with graphics, buttons, charts and forms instead of plain text. The format adapts to the question: comparisons show up side by side, and explanations can turn into interactive charts. Users can also have small tools built inside the chat, such as a savings calculator or a game.

OpenAI's own post on X promises "fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot." That is a description of a feeling. It is not a measurement.

Google got there first, at least in name. The Decoder notes that Google shipped a similar feature to Gemini in May, called "Neural Expressive."

What the announcement does not include: any account of what powers the mini-apps. The languages, frameworks and code runtime behind those calculators and games are not stated.

The numbers, such as they are

GPT-6 can also respond while still "thinking." OpenAI says this cuts wait times by 44 percent and can produce better answers. The reporting does not say what the 44 percent is measured against, on which tasks, or whether it is a median, an average or a best case. A percentage with no baseline is a decoration.

The quality claim is similarly modest. In internal tests, according to The Decoder, GPT-6 scored higher than its predecessor GPT-5.6 on difficult web searches using this method. No score, benchmark name or sample size accompanies it. "Scored higher" is true of a one-point gap and of a twenty-point gap alike. It also covers one method on one task type, and says nothing about how Sol and Luna compare.

That comparison matters, because the tiers are the product. The material available gives no technical differences between GPT-6 Sol and GPT-6 Luna, and no pricing or rate limits for Free, Go, Plus, Pro, Business or Enterprise.

A smaller puzzle: a post by @btibor91 on X carries the headline "GPT‑6 Instant and Intelligent UI in ChatGPT." What GPT-6 Instant is, and how it relates to Sol or Luna, is not explained anywhere in the material.

Everyone, but not all at once

OpenAI's post and its blog headline say GPT-6 and Intelligent UI are for everyone. The Decoder gives the schedule: the rollout starts globally today for Plus, Pro, Business and Enterprise users, and Free and Go users follow one day later. Both can be true. "Everyone" is a destination, and tomorrow is when the free tier gets there.

There is also a timeline wrinkle. In a statement carried by Techmeme, OpenAI said: "Last month we introduced the first GPT-6 models for paid customers and today we're bringing that next generation of intelligence to more people …" That implies paid customers already had GPT-6 models a month ago, while The Decoder describes today's rollout as starting with paid tiers. The material does not reconcile the two, and it does not say whether today's paid-tier models differ from last month's.

For the record on enthusiasm: OpenAI's announcement post on X had 3,021 likes and 153,798 views at last count. Engagement is not an eval, but it is an eval-shaped number, which is how these things get through.

What it means

The real change here is not the model, it is the format. A text answer that is wrong is wrong in a sentence you can interrogate. A chart that is wrong is wrong with axes, color and a slider, which is a more persuasive way to be wrong. A savings calculator built inside a chat is only as good as its arithmetic, and the cited reporting provides no test of that.

That is not an argument against the feature. Side-by-side comparisons beat paragraph-by-paragraph ones, and a form that does what a form should is a fine thing. It is an argument for measuring the interactive output separately from the prose, because the material gives no accuracy figure for either.

What to watch

• Tomorrow's Free and Go rollout, and whether Luna users see the same Intelligent UI as Sol users.
• Any published numbers behind the 44 percent and the web-search result: baseline, benchmark, sample.
• An explanation of Sol versus Luna, and of whatever GPT-6 Instant turns out to be.
• Pricing and rate limits by tier.
• How Intelligent UI holds up against Google's May feature once someone runs the two on the same questions.

Show me the eval, not the vibe. Until then, the buttons are very nice.