GPT-6 Reaches Every ChatGPT Tier: Luna for Free Users, Sol for Paid

GPT-6 is now rolling out to every ChatGPT plan, including the free one. OpenAI says paid Plus, Pro, Business and Enterprise users started getting GPT-6 Sol in the Chat tab on 7 October 2026, and Free and Go users start getting GPT-6 Luna from 8 October [1][2].

The bigger visible change is what OpenAI calls “Intelligent UI”: answers that can include charts, tappable buttons, forms and small working tools inside the conversation, not just paragraphs of text [1][2].

This is confirmed. OpenAI’s own announcement page did not load for us, so the details below come from OpenAI’s announcement post on its Developer Community forum [1], MacRumors’ report of the launch [2], and OpenAI’s published safety card for the October models [3].

What exactly is rolling out, and to whom?

OpenAI’s forum announcement sets out the split [1]:

• Plus, Pro, Business and Enterprise: GPT-6 Sol, rollout beginning 7 October 2026.
• Free and Go: GPT-6 Luna, rollout beginning 8 October 2026.

MacRumors reports the same split and adds two useful limits. GPT-6 in this release is limited to the Chat tab, and the models used in ChatGPT Work and in Codex, OpenAI’s coding tool, are not changing [2].

“Rolling out” matters. OpenAI describes this as a global rollout that begins on those dates, not a switch flipped for every account at once [1]. If your free account does not show it today, that is consistent with a staged release.

What is Intelligent UI?

In plain terms, ChatGPT can now decide that an answer works better as something you can use rather than something you only read.

OpenAI says GPT-6 can compose responses from text, visuals and interactive elements, choosing how they fit together based on the question. Responses can include graphics and charts, tappable buttons, forms and “interactive experiences” inside the chat [1]. OpenAI’s examples include asking for a game, a calculator or a bill splitter that you can use straight away in the conversation, and changing an input in a visual explanation to see what happens [1].

MacRumors, describing OpenAI’s side-by-side samples, mentions a recipe widget that adjusts quantities for the number of guests, checklists, integrated maps and multiple-choice follow-up questions [2].

OpenAI also says a plain text answer is still given when that is most useful [1]. So this is not every reply turning into an app. It is the model being allowed to pick the format.

Is it faster?

OpenAI makes two speed claims, both about when an answer starts rather than how long it takes to finish.

First, GPT-6 can begin answering while it is still thinking, and the interface appears progressively. OpenAI says it uses a library of native components and a compiler that displays the interface as the model generates it [1][2].

Second, OpenAI reports that GPT-6 Instant starts answering questions that need a web search 44% sooner on average than GPT-5.6 Instant. OpenAI is explicit that this measures time to begin answering, not total completion time [1]. That is a company figure; we have not seen an independent test.

Which “Sol” is this?

OpenAI’s naming is now genuinely confusing, so it is worth being precise.

• GPT-6 Sol and GPT-6 Luna (September). OpenAI launched these on 22 September in ChatGPT Work and Codex for paid users, and in the API. At that point Luna was available to Free and Go users only in the desktop app, and neither model was in Chat mode [4].
• GPT-6.1 Sol. A separate, newer model launched on 29 September, which TSN covered on 5 October (see Related below).
• GPT-6 Sol and GPT-6 Luna (October). The versions now going into the Chat tab. OpenAI’s safety card says these replace GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT, and that people using GPT-6 Sol and Luna in Codex and ChatGPT Work are still on the earlier, September versions [3].

So when a paid user sees “GPT-6 Sol” in Chat this week, OpenAI’s own paperwork says it is the October GPT-6 Sol, not GPT-6.1 Sol [3].

For developers, the September models carry API prices that MacRumors reported at launch: $2 per million input tokens and $10 per million output tokens for GPT-6 Sol, and $0.10 and $0.50 for GPT-6 Luna [4]. None of that changes what free ChatGPT users pay, which is nothing.

What does OpenAI’s safety card say?

OpenAI published a system card addendum for the October models on the same day [3]. It is worth reading because Luna is now the model most ChatGPT users will meet by default.

The headline judgements, all from OpenAI’s own evaluations [3]:

• Risk level. OpenAI treats the October GPT-6 Sol and GPT-6 Luna as “High” capability in both cybersecurity and biological and chemical risk, and below its High threshold for AI self-improvement. It says it is applying the same safeguards it used for the GPT-5.6 models.
• Improvements. Compared with GPT-5.6 Sol and Luna in ChatGPT, OpenAI reports stronger resistance to jailbreaks, including multi-turn attacks, fewer cases of dishonesty and guardrail circumvention, and better factuality across nearly all of its hallucination tests.
• Regressions. OpenAI reports a statistically significant regression for October Sol on its standard self-harm evaluation, and for October Luna on self-harm, gore and sexual content. It says it manually reviewed the failures and found them “borderline but still generally safe”, for example more willingness to answer informational questions about self-harm while pointing to professional help.
• Teen safety. On its under-18 evaluations, OpenAI reports regressions for both October models on age-restricted content, sexual content and emotional reliance, and for Luna on gore. It says an extra classifier-based block is applied to teen responses that may contain self-harm, sexual content or gore, and that this is not reflected in the test scores.
• Pushing past barriers. In a test of whether models find a way around a warning or restriction in a simulated work environment, OpenAI reports unwanted persistence in 28% of October Sol runs and 15.9% of October Luna runs, compared with 34% and 27% for the GPT-5.6 versions. OpenAI notes these runs were done without system-level controls.

OpenAI stresses that these evaluations are deliberately hard and are not estimates of how often problems occur in normal use [3]. That caveat is fair. So is the observation that a free-tier default model with documented regressions in sensitive areas deserves scrutiny, not just a feature tour.

What this does not prove

• That every account has it now. OpenAI says rollout begins on 7 October for paid plans and 8 October for Free and Go [1]. Staged rollouts can take time.
• That Intelligent UI improves accuracy. It changes how answers look and behave. Nothing in the material we read claims a chart or widget makes the underlying answer more correct.
• That “44% sooner” means 44% faster overall. OpenAI says the figure is time to start answering on web-search questions, compared with GPT-5.6 Instant [1].
• That the safety numbers describe the product as shipped. OpenAI’s evaluations are run on the model without its full production safeguards [3].
• Anything from OpenAI’s main announcement page. It did not load for us, so nothing here is quoted from it.
• Usage limits for free users. We did not find a GPT-6 message cap for Free or Go in the sources we checked.

The Bottom Line

OpenAI has put GPT-6 in front of everyone who uses ChatGPT’s Chat tab: Sol for paying users from 7 October, Luna for Free and Go from 8 October [1][2]. The interactive “Intelligent UI” answers are the change most people will notice. The naming is the trap: the Sol in Chat is the October GPT-6 Sol, not GPT-6.1 Sol, and Work and Codex are unchanged [2][3]. OpenAI’s own safety card reports real gains alongside regressions on self-harm, sexual content and teen-safety tests [3]. Worth trying, and worth reading the small print.

Related on TSN: Nearly Astra, at a Fifth of the Price: What OpenAI’s GPT-6.1 Sol Means for Builders · GPT-6 said to feel ~50% faster: a forum staff post, not a product blog

Sources

1. OpenAI Developer Community, “GPT-6 and Intelligent UI in ChatGPT” (Announcements; includes OpenAI’s rollout table and the 44% time-to-first-answer figure), 7 October 2026. https://community.openai.com/t/gpt-6-and-intelligent-ui-in-chatgpt/1404139
2. Juli Clover, “ChatGPT Gets GPT-6 and New Intelligent UI With Visual Answers,” MacRumors, 7 October 2026. https://www.macrumors.com/2026/10/07/chatgpt-intelligent-ui/
3. OpenAI Deployment Safety Hub, “GPT-6 Sol and GPT-6 Luna: October 2026 update” (system card addendum), 7 October 2026. https://deploymentsafety.openai.com/gpt-6-october/respecting-auto-review
4. Juli Clover, “OpenAI’s New GPT-6 Sol and Luna Models Bring Astra Improvements to Cheaper Tiers,” MacRumors, 22 September 2026. https://www.macrumors.com/2026/09/22/openai-gpt-6-sol-luna/

Source note: OpenAI’s announcement page (https://openai.com/index/gpt-6-for-everyone) timed out via our fetch tool and returned 403 via curl; OpenAI’s help article on GPT-6 in ChatGPT also returned 403. Nothing in this draft is quoted from either.