ChatGPT GPT-6 and Intelligent UI: Rollout by Plan

Which plan got which model

On Wednesday, October 7, 2026, OpenAI began rolling out GPT-6 with Intelligent UI to the Chat tab for ChatGPT Plus, Pro, Business and Enterprise users worldwide, and said the rollout would reach the Free and Go tiers the following day. Paid plans run on GPT-6 Sol and Free and Go on GPT-6 Luna, both tuned for everyday conversation. Enterprise availability depends on admin settings. [1][2]

The change applied only to Chat. OpenAI said the models behind ChatGPT's Work and Codex surfaces were not changing in this release. Sol and Luna themselves were not new: OpenAI had released them on September 22, which OpenAI's October post described as introducing the first GPT-6 models for paid customers. Its before-and-after examples compare GPT-6 Instant with GPT-5.6 Instant. OpenAI framed this release as bringing GPT-6 to the more than 1.2 billion people it said use ChatGPT each week. [1]

How Intelligent UI works

Intelligent UI lets an answer mix text with graphics, tappable buttons, forms, charts and interactive tools, with the layout chosen per question: a comparison side by side, an explanation as an interactive diagram, a road trip on a map, or a savings calculator or bill splitter built on request. When plain text works best, ChatGPT is meant to answer in plain text. [1][2]

OpenAI described two pieces. A library of native, streamable components gives every response a common design base, and a compiler renders the interface while the model is still generating, so it appears progressively instead of all at once. OpenAI said it expanded its training to grade the interfaces the model creates for clarity, usefulness and completeness, so that GPT-6 learned when interactivity helps and when it does not. TechCrunch reported that users will be able to dial the visuals back. [1][2]

The design choice is a constrained one. Rather than letting the model write arbitrary web pages, OpenAI gave it a fixed kit of parts and trained its judgment about when to use them, which keeps answers fast and consistent across web and mobile at the cost of what the model can invent. OpenAI said itself that the model's design judgment still needed work. [1]

Answering while it thinks

OpenAI also trained GPT-6 in ChatGPT to start answering before it finishes reasoning, building the reply across several partial responses with the user's waiting time in mind. It reported that on an internal set of everyday agentic tasks, GPT-6 at Extra High effort starts answering as quickly as GPT-5.6 at Medium while scoring better overall than GPT-5.6 at Extra High, and that GPT-6 Instant starts answering web-search questions 44 percent sooner on average than GPT-5.6 Instant. Both figures come from OpenAI's internal evaluations. [1]

Safety claims

OpenAI said GPT-6 builds on safety work from GPT-6 Astra to resist attempts to bypass its training, particularly attacks that adapt over several turns, and that its safety training was updated against high-risk misuse involving cyberattacks, biological threats and violence, using conversation history to spot risks a single prompt would not show. It said the model is better at saying when it lacks the information or tools to answer, and pointed to a system card for the results. [1]