GPT-6 Intelligent UI is now available to free ChatGPT users. OpenAI began the rollout on 7 October 2026 for Plus, Pro, Business and Enterprise accounts, then extended it to Free and Go accounts on 8 October. Answers can now include charts, forms, buttons and small in-chat tools.
What Intelligent UI Actually Changes in an Answer
Replies stop being a single block of text and become whatever shape the model picks. OpenAI, the San Francisco company behind ChatGPT, says the model chooses between prose, a side-by-side comparison, a diagram, a map or a working tool, and still returns plain text when that serves the question better.
The examples OpenAI gave are deliberately mundane: a cooking timeline shown beside a recipe, a road trip with stops plotted on a map, a savings calculator with inputs the reader can adjust, a tool for splitting a restaurant bill, and a small game played inside the conversation. A user can also ask for an interface directly rather than waiting for the model to offer one.
Underneath, OpenAI describes a library of native streamable components paired with a compiler that builds the interface while the model is still writing, so parts of an answer appear before the whole response is finished. The same release lets the model start answering a complex question while it continues to reason or call tools, adding to the reply as it goes.
Which Plan Gets Which Model
Two different models sit behind the same feature, split by what the account pays. OpenAI had already released both to its developer API and to ChatGPT Work and Codex on 22 September 2026; this week’s change brings them into the main Chat tab.
| Plan | Model in Chat | Intelligent UI from |
|---|---|---|
| Plus, Pro, Business, Enterprise | GPT-6 Sol | 7 October 2026 |
| Free and Go | GPT-6 Luna | 8 October 2026 |
| Pro thinking level | GPT-6 Astra | Not supported |
Enterprise accounts are a separate case: the feature arrives only once a workplace administrator switches it on, so employees on an Enterprise plan may still see nothing. In the model menu, GPT-6 takes over the slot previously labelled “Latest”, while GPT-5.6 Sol stays selectable as its own option. ChatGPT Work and Codex keep the models they were already running.
The gap between the two tiers is not cosmetic. On OpenAI’s developer pricing, GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, against $0.10 and $0.50 for GPT-6 Luna on prompts up to 272,000 input tokens. Free users are getting the interactive format on the cheaper model. Rival free tiers have been moving the other way, with Google tightening Gemini model limits for free users from 9 October.
Where Intelligent UI Does Not Work
Three exclusions apply, and two of them will catch paying customers by surprise.
- The Pro thinking level: it runs on GPT-6 Astra, which does not support Intelligent UI. The GPT-6 Astra model launched in September 2026 as the family’s most capable tier.
- Older desktop apps: the previous ChatGPT applications for macOS and Windows cannot render it, so the browser or an updated mobile app is required.
- Thinking levels above Extra High: support runs from Instant to Extra High only.
As of 9 October 2026, the rollout is described by OpenAI as gradual across the web and updated apps, so an eligible account may not see the feature immediately. OpenAI’s announcement says the release reaches more than 1.2 billion weekly ChatGPT users.
How to Reduce the Visuals
There is a setting, but it dials the interface down rather than switching it off. OpenAI’s help pages describe it as a web-only control, and the company is explicit that some visual and interactive elements can still appear afterwards.
- Open ChatGPT in a web browser and go to Settings.
- Choose Personalization.
- Select Layout and Visuals.
- Set it to Simple.
No equivalent control has been documented for the iOS or Android apps. The setting also governs appearance rather than the model, and changing how one conversation looks does not change the others. Readers who have hunted for opt-outs in ChatGPT before will recognise the pattern: when visual ads arrived in the United States in October 2026, opting out carried its own cost.
The Speed Claim, and What It Measures
OpenAI’s headline performance number is narrower than it sounds. The company says GPT-6 Instant starts answering web-search questions 44 percent sooner on average than GPT-5.6 Instant — a measure of how quickly the first words appear, not of how long a complete answer takes.
A second internal result compares reasoning tiers: OpenAI reports that GPT-6 Extra High began answering roughly as fast as GPT-5.6 Medium while scoring higher overall than GPT-5.6 Extra High on what it calls high-value everyday agentic tasks. OpenAI has not published the underlying task set or sample sizes for either claim, and both come from its own evaluations rather than an outside benchmark.
On search behaviour, OpenAI says the model judges more accurately when a lookup is worth doing and more reliably finds material that supports its answer. It gave no figure for either. The company also states plainly that the model’s design judgment and the range of interfaces it can build still need work.
What OpenAI’s Own Safety Update Reports
Alongside the launch, OpenAI published a safety update that records several measurements moving in the wrong direction. The document is dated 7 October 2026 and covers the October versions of GPT-6 Sol and GPT-6 Luna.
Under its Preparedness Framework, OpenAI says it is treating both models as High capability in the Cybersecurity domain, though below Critical, and as High capability in the Biological and Chemical domain. Neither model reaches its High threshold in AI Self-Improvement.
The regressions are measured against the GPT-5.6 models. On OpenAI’s production benchmarks, GPT-6 Sol shows a statistically significant regression on standard self-harm, and GPT-6 Luna shows them on standard self-harm, gore and sexual content. In evaluations covering users under 18, both models regress on age-restricted content, sexual content and emotional reliance, with gore added for Luna. Both also regress on an extremism evaluation in the vision tests.
OpenAI’s characterisation of those results matters as much as the results. It describes the production violations as “borderline but still generally safe”, says the Luna gore and sexual content outputs were of “lower severity”, and reports that manual review and adversarial red teaming found the disallowed responses to be generally low severity, with no high-severity risks identified against its evaluation criteria. On a different axis the same document reports gains, calling both models “highly robust to prompt injection” and saying they saturate instruction hierarchy evaluations at 99.99 percent for Sol and 99.79 percent for Luna.
Two further cautions come from reporting rather than from OpenAI. Gizmodo noted that richer visual answers may consume a user’s tokens faster, and that generated interfaces create an additional surface on which a model can present something incorrect. It also pointed out that Google shipped a comparable Generative Interface feature with Gemini 3 without obviously pulling ahead of its rivals.
Frequently Asked Questions
Is GPT-6 Intelligent UI Free?
Yes. Free and Go accounts received Intelligent UI on 8 October 2026, running on the GPT-6 Luna model. Paid plans received it a day earlier on GPT-6 Sol.
Why Can I Not See Intelligent UI?
Four reasons are documented: the rollout is gradual, the Pro thinking level does not support it, the older macOS and Windows desktop apps cannot render it, and Enterprise accounts need an administrator to enable it.
Can I Turn Off Intelligent UI?
Not completely. On the web, Settings then Personalization then Layout and Visuals set to Simple reduces visual and interactive elements, but OpenAI says some may still appear. No equivalent setting has been documented for the mobile apps.
What Is the Difference Between GPT-6 Sol and GPT-6 Luna?
Sol is the model given to Plus, Pro, Business and Enterprise accounts and Luna the one given to Free and Go. On OpenAI’s developer pricing Sol costs $2 per million input tokens against $0.10 for Luna, indicating a substantial capability gap.
Does GPT-6 Make ChatGPT Faster?
For one specific case. OpenAI says GPT-6 Instant begins answering web-search questions 44 percent sooner on average than GPT-5.6 Instant, which measures time to the first words rather than total response time.
Did OpenAI Report Any Safety Problems With GPT-6?
Yes. Its 7 October 2026 safety update lists statistically significant regressions against GPT-5.6 on standard self-harm for both models, plus gore and sexual content for Luna, and several under-18 categories for both. OpenAI describes the responses involved as generally low severity.
