For years, a ChatGPT reply was a block of text. On October 7, 2026, OpenAI began giving paying customers a GPT-6 update it calls Intelligent UI, and it scheduled free accounts for the next day on a different model. A Wednesday headline can make that sound like one switch for every login. The plan name, and whether you are in the chat tab, decide what shows up.
TechCrunch, MacRumors, and The Indian Express describe the same kind of answer. A reply can still be ordinary text. It can also mix in charts, tappable buttons, forms, and small tools such as a calculator or a bill splitter. TechCrunch's Lucas Ropek, in a story timestamped 11:00 a.m. Pacific on October 7, wrote that OpenAI staff walked journalists through the feature. Product manager Aarush Selvan said ChatGPT has "predominantly been a text-based interface." The example he used was an airplane wing: ask how a wing generates lift, and the reply draws diagrams rather than leaving the whole explanation as paragraphs. Ropek also wrote that people who want fewer visuals can dial them back.
MacRumors and a same-day post on OpenAI's developer community describe how the screen is built. Intelligent UI uses a library of native components and a compiler that draws the interface while the model is still writing it. GPT-6 can begin the answer while it keeps thinking or using tools. The Indian Express story, timestamped the morning of October 8 in India, quotes OpenAI's announcement for that behavior and reports the model split: paid ChatGPT plans use GPT-6 Sol, and Free and Go use GPT-6 Luna. MacRumors reports the same split.
Who gets which model, and when
The community post prints the rollout as a table. MacRumors and The Indian Express match it. TechCrunch names the same days and the same groups of plans, without the Sol and Luna names.
| Plans | When the rollout starts | Model in Chat |
|---|---|---|
| Plus, Pro, Business, and Enterprise | October 7, 2026 | GPT-6 Sol |
| Free and Go | October 8, 2026 | GPT-6 Luna |
That is a one-day gap, and it is also a model gap. October 7 access and October 8 access are not the same GPT-6. Sol is the paid-plan model in this chat release. Luna is the Free and Go model.
MacRumors and The Indian Express both say the release leaves ChatGPT Work and Codex on the models those products already used. Someone working in the chat tab is inside this rollout. Someone whose coding or work product sits in Work or Codex is outside it.
What a 44 percent faster start leaves in place
The community post says OpenAI reports that GPT-6 Instant starts answering questions that need web search 44 percent sooner, on average, than GPT-5.6 Instant. The post says this measures time to begin answering, not time to finish. The Indian Express reports the same 44 percent figure, tied to the same situation: questions where a web search is required, and the clock that matters is the start of the reply.
Neither account publishes the underlying number of seconds. The percentage still converts cleanly. A start time that is 44 percent shorter keeps 56 percent of the old start time, because 1 minus 0.44 equals 0.56. On a base of 100, the new start is 56. A wait of 100 units of time before the first part of the answer would become 56 units under OpenAI's claim. Geeknewz has not timed either model. The figure says nothing about how long the finished answer takes after it has started.
The Indian Express also passed along a second internal comparison. On tasks OpenAI described as high-value everyday agentic work, GPT-6 Extra High began responding in roughly the same amount of time as GPT-5.6 Medium, and it scored higher overall than GPT-5.6 Extra High. That comparison appears in one of the reports read for this story. It is OpenAI's description of its own evaluation, and the outlet did not print a score table a reader can recompute.
The same story says the company claims more than 1.2 billion people use ChatGPT each week. That number is about the size of the audience. It is not a count of how many of those people had Sol or Luna on the afternoon of October 7.
Verdict: Check the plan, then the tab. Plus, Pro, Business, and Enterprise were the October 7 group, on GPT-6 Sol. Free and Go were the October 8 group, on GPT-6 Luna. TechCrunch says the extra visuals can be turned down. The speed claim is a start-time claim, equal to 56 percent of the previous wait, and it applies to questions that need web search. Work and Codex stay on the models they already had.
Photo: Shixart1985 / Wikimedia Commons, CC BY 2.0.
