GPT-6 Comes to ChatGPT's Chat Tab - Sol on Plus and Above, Luna on Free/Go, Plus Intelligent UI That Puts Diagrams and Buttons in Answers

On October 7, 2026 (US time), OpenAI began deploying GPT-6 in the Chat tab of ChatGPT¹. The rollout covers Plus, Pro, Business and Enterprise from that day, with Free and Go following from October 8. Alongside the model, two changes arrive: “Intelligent UI,” which lets an answer contain diagrams, buttons, forms, charts and similar elements, and a way for the model to start replying before it has finished thinking.

OpenAI shipped its first GPT-6 models to paying users last month. But when GPT-6 Sol and Luna launched on September 22, they went to ChatGPT Work and Codex, not to Chat. GPT-6.1 Sol on September 29 likewise started in Work and Codex. This release targets Chat itself, so the model behind everyday conversations changes, Free and Go included.

Which Model Runs Depends on the Plan

In Chat, Plus, Pro, Business and Enterprise run GPT-6 Sol, while Free and Go run GPT-6 Luna. OpenAI says it adjusted both for ordinary conversation¹. The release notes add exceptions that the announcement page leaves out².

Plan	Model in Chat	Rollout start (US time)	Notes
Plus, Pro, Business	GPT-6 Sol	October 7	Pro’s reasoning option remains on GPT-6 Astra
Enterprise	GPT-6 Sol	October 7	Depends on workspace admin settings
Free, Go	GPT-6 Luna	October 8

In terms of reasoning levels, Intelligent UI works from Instant up to Extra High. Pro’s reasoning option still runs on GPT-6 Astra and has no Intelligent UI support². Put those two statements together, and a Pro user who wants answers with diagrams or interactive elements has to pick one of the Instant-to-Extra High levels rather than the reasoning option.

Both the announcement and the release notes state that the change is limited to Chat and that Work and Codex keep their current models¹². The model added to Chat is GPT-6 Sol, not GPT-6.1 Sol, whose deployment is under way on the Work side.

Edu does not appear in the list of plans covered by this release. Enterprise is also the only plan whose availability is explicitly tied to admin settings; nothing is said about Business in that respect.

Intelligent UI Picks a Format for Each Question

According to OpenAI, GPT-6 was trained to build answers out of text, visuals and interactive pieces, and to decide how to combine them depending on what was asked¹. The company’s examples: a side-by-side layout may suit a comparison, and an interactive diagram may suit an explanation. When plain text is the most helpful answer, it answers in text. On request, it can also build a small tool inside the conversation for whatever task the user is working on; OpenAI’s examples include a savings-growth calculator and a tool for dividing a restaurant bill.

Under the hood, OpenAI says it created a set of streamable, platform-native UI components, plus a compiler that handles the interface while the model is still generating it. As a result, the screen assembles itself piece by piece instead of waiting for the full answer. The library supplies a shared design base, and the model decides how to arrange the parts; OpenAI says this keeps the experience quick and familiar on both web and mobile.

The company acknowledges that the model’s design choices and the range of things it can build still need improvement. TechCrunch reports that users who find the visuals excessive will be able to reduce them, in the same way as other ChatGPT personality settings³. This control, however, is not mentioned on the announcement page or in the release notes.

Answers That Start Mid-Thought

The other change concerns how responses arrive. GPT-6 can start writing an answer while it is still reasoning or calling tools, then append what it finds later without the user sending another prompt². Users can begin reading before the whole response is complete.

On speed, OpenAI points to internal evaluations. In an internal test of everyday agentic tasks, GPT-6 at Extra High reportedly started answering as fast as GPT-5.6 at Medium while scoring higher overall than GPT-5.6 at Extra High¹. For questions requiring web search, OpenAI says GPT-6 Instant on average gets its reply started 44% earlier than GPT-5.6 Instant does. On search-backed answer quality, the company says that in an internal set of hard problems, GPT-6 got the core of the question right more often than GPT-5.6. All of these figures come from OpenAI’s own evaluations.

On safety, the company writes that adversarial testing showed GPT-6 holding up better against attempts to get around its safety training, especially attacks that shift tactics over several turns. Details are left to the system card.

What Changes for Free and Go Users and for Administrators

According to OpenAI, ChatGPT’s weekly user count exceeds 1.2 billion. For people on Free or Go, Chat answers will now come from GPT-6 Luna and will start to include diagrams and interactive elements.

For business use, two points are worth checking. First, on Enterprise, availability depends on admin settings, so when employees see the new model is decided by the company’s own configuration as well as OpenAI’s rollout schedule. Second, this change leaves the Work and Codex models untouched. Even though Chat answers now look different, this announcement does not alter the assumptions behind work built in Work or Codex.

Other vendors are also splitting model access by plan. Google is dividing Gemini app model access by plan, and from October 9 users without a subscription can choose only Flash-Lite. In ChatGPT’s Chat, Free and Go now get Luna, Plus, Pro, Business and Enterprise get Sol, and Pro’s reasoning option uses Astra. “GPT-6 in ChatGPT” therefore means different things depending on the plan and the tab, which is worth keeping in mind when writing internal guidance on how to use it.

Sources

1. GPT-6 and Intelligent UI for everyone - OpenAI official announcement (October 7, 2026)
2. ChatGPT release notes - OpenAI Help Center (entry “GPT-6 and Intelligent UI in ChatGPT” dated October 7, 2026; accessed October 8, 2026)
3. ChatGPT is getting a lot more visual, with the launch of a new interface - TechCrunch (October 7, 2026)