Updated October 7, 2026. ChatGPT’s next big change is about what an answer can become. With GPT-6 and Intelligent UI, a conversation can contain an interface you actually use: a comparison, an explanation with controls, or a small tool built around your question.
The rollout begins October 7 for Plus, Pro, Business and Enterprise, with Free and Go following from October 8. Paid Chat uses GPT-6 Sol; Free and Go use GPT-6 Luna as the update reaches them. This release concerns the Chat tab; Work and Codex retain their existing models. These are rollout start dates, not a promise that every account switches simultaneously. Source: OpenAI’s October 7 release notes.
Our assessment: the interesting advance is the shorter distance between asking a question and working with the answer. A paragraph can explain a variable. A control lets you change it and investigate the consequences.
What is Intelligent UI?
Intelligent UI lets GPT-6 combine prose, graphics, buttons, forms and charts into a response suited to the question. OpenAI built native components that can stream into view, plus a compiler that processes the interface during generation. The model has also been trained to choose layouts and interactions. Source: the launch announcement.
Think of an answer that has useful controls attached. If you are comparing options, you want to inspect the differences. If you are learning a concept, you want to test your understanding. If you are calculating something repeatedly, you want to change the inputs without rewriting the same request.
This is why the feature matters even to people who never write code. The value is in describing the outcome you need and receiving something you can operate. Whether that saves time depends on the task: a one-sentence factual question still deserves a one-sentence answer.
What does it look like? Watch OpenAI’s launch video
Official OpenAI video embedded from the launch page. If the player is unavailable, the source page also contains the demos.
Official interactive demonstration hosted by OpenAI. Scroll inside the example to explore its controls. This is a launch demo, not your private ChatGPT conversation.
A working example: the bill splitter
OpenAI’s public bill-splitter demo resembles a clipboard. It has fields for five diners, editable dishes and prices, checkboxes for who shared each item, tax controls and tip presets. Its sample receipt totals $107.78 before an added tip.
That gives a concrete way to judge the idea. Assigning a shared dish to two people should divide its cost between those people; a tip control should update the total consistently. The quality of this experience depends on those relationships being right, as well as the interface being readable.
Open and try the official interactive demo →
We inspected OpenAI’s public demo and documentation for this guide. This is launch analysis, not a claim that we independently benchmarked the production ChatGPT rollout.
What is it for?
Our practical test is simple: would changing an input, comparing alternatives or manipulating a diagram make the answer more useful? If so, an interactive response is worth trying. Here are original prompt ideas to adapt to your own work.
| Task | Prompt to try | What to check |
|---|---|---|
| Understand a concept | “Explain compound growth with a chart and controls for starting amount, contribution and growth rate. Show the assumptions.” | Changing one input should produce a coherent, explainable change. |
| Compare products | “Compare these three laptops using my supplied specifications. Let me sort by weight, price and battery estimate; label missing data.” | Each specification should trace back to the supplied information. |
| Plan a project | “Turn this event brief into a timeline I can explore by phase. Highlight dependencies and unresolved decisions.” | Dates, dependencies and responsibilities should agree with the brief. |
| Explore a budget | “Make a small budget tool using these categories. Let me adjust each amount and show the remaining balance.” | Totals should reconcile, including zero and unusually large inputs. |
| Study actively | “Create a short practice experience from these notes. Let me reveal explanations after choosing an answer.” | Feedback should explain the answer instead of simply marking it correct. |
These prompts are suggestions, not guaranteed templates. Start with your goal and the information that must remain accurate. Then specify which inputs you want to change and what feedback would help you make sense of the result.
For creators and educators, the opportunity is an explanation that invites experimentation. For everyday users, it is avoiding repeated back-and-forth over the same calculation. For teams, it is making a decision easier to inspect. A polished presentation alone does not establish that the underlying recommendation is sound.
Who can use it, and when?
| ChatGPT plan | Rollout starts | Model for this Chat update |
|---|---|---|
| Plus, Pro, Business, Enterprise | October 7, 2026 | GPT-6 Sol |
| Free, Go | October 8, 2026 | GPT-6 Luna |
The rollout is global; Enterprise access also depends on admin settings. Intelligent UI supports Instant through Extra High where your plan offers those levels. Pro the subscription and Pro the reasoning setting are different things: the Pro reasoning option uses Astra and does not support Intelligent UI. Source: release notes.
Use the web or supported updated apps. OpenAI says the older macOS and Windows desktop apps do not support these capabilities. Plus includes Medium and High; Pro, Business and Enterprise also include Extra High. Actual account and workspace permissions still apply. Source: model-access guidance.
Launch-day documentation caveat: the Help Center still contains GPT-5.6 Luna language for Free and Go. Today’s release notes explicitly announce GPT-6 Luna for their October 8 rollout. We use the dated announcement for that schedule; do not interpret the older wording as confirmation that your account has already switched.
What does GPT-6 with Intelligent UI cost?
There is no separate Intelligent UI usage quota; existing model, tool and plan allowances apply. OpenAI has not announced a separate Intelligent UI add-on fee. Source: Intelligent UI help.
| Plan | Published subscription price | Pricing context |
|---|---|---|
| Free | $0 | Free-tier allowances apply. |
| Go | $8/month U.S. reference price | Local pricing varies; check your account’s current checkout. |
| Plus | US$20/month | Monthly individual subscription. |
| Pro 100 / 200 / 500 | US$100 / $200 / $500 per month | Different included usage allowances. |
| Business Standard | US$25 per user/month, or $20 with annual billing | Minimum two paid seats across Standard and Premium. |
| Business Premium | US$125 per user/month, or $100 with annual billing | Higher-capacity business seat. |
| Enterprise | Contact sales | Contract-specific pricing and access. |
Price sources: ChatGPT pricing, Go’s published U.S. launch price, Plus, current Pro tiers and Business FAQ. Local currency, taxes and purchasing channel can change the amount you pay. The Go figure is a published reference, not a newly announced October price.
Our buying advice: try the access already included in your plan before upgrading for this feature. A higher subscription can make sense for broader capabilities or higher allowances, but the rollout itself includes Free. Also, the $500 Pro tier is not an Intelligent UI requirement. Its distinctive Astra Ultrafast access concerns Work and Codex, according to OpenAI’s Pro guide.
Specs: distinguish the Chat product from the API models
OpenAI’s October system card explicitly distinguishes October’s Chat models from the September versions still used in Work and Codex. Treat a shared family name as a starting point, not proof that every product has identical behavior or limits. Source: October system card.
The following are published API model specifications. They are useful developer context, but they do not establish a million-token context allowance for your ChatGPT subscription or expose ChatGPT’s Intelligent UI renderer through a normal API call.
| API specification | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| Model ID | gpt-6-sol | gpt-6-luna |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Maximum output | 128,000 tokens | 128,000 tokens |
| Knowledge cutoff | April 20, 2026 | May 18, 2026 |
| Native modalities | Text input/output; image input | Text input/output; image input |
| Standard input / output per million tokens | US$2 / $10 | US$0.10 / $0.50 |
| Cache read / write per million tokens | US$0.20 / $2.50 | US$0.01 / $0.125 |
Sources: Sol model card and Luna model card. Both support streaming, function calling and structured outputs. The listed models do not natively output audio or video; supported tools are separate capabilities. Requests above 272,000 input tokens carry higher rates, and service tiers, regional processing and tools can affect costs.
API usage is billed separately from Plus and Go subscriptions. A ChatGPT plan price therefore cannot be compared directly with a per-million-token rate as though they purchase the same thing. Source: Plus subscription details.
The launch materials reviewed for this article do not establish a public parameter count, model-weight download, local GPU requirement or universal Chat context limit for the October experience. Those should remain unknown rather than be filled in using another GPT-6 model’s specifications.
GPT-6 can also start replying while it continues reasoning or using tools, then add findings without requiring another prompt. Source: release notes.
Benchmarks: faster first answers, with important limits
OpenAI reports that GPT-6 Instant begins answering web-search questions 44% sooner on average than GPT-5.6 Instant. In an internal everyday-agentic evaluation, GPT-6 Extra High began answering as quickly as GPT-5.6 Medium while scoring above GPT-5.6 Extra High. Source: launch evaluation claims.
That is a first-response latency claim. It does not mean every task finishes 44% faster, or that every interactive screen appears instantly. For a user, both the first useful response and the time to a correct, complete result matter. We have not independently reproduced these measurements.
October-model results from the system card
| Evaluation | GPT-5.6 Sol, August | GPT-6 Sol, October | GPT-5.6 Luna, August | GPT-6 Luna, October |
|---|---|---|---|---|
| HealthBench Professional, length-adjusted | 54.0 | 55.2 | 44.1 | 48.2 |
| HealthBench, length-adjusted | 55.0 | 52.1 | 53.3 | 50.4 |
| HealthBench Hard, length-adjusted | 31.4 | 27.8 | 28.7 | 27.2 |
| MentalHealthBench, overall | 47.53 | 48.81 | 45.11 | 51.73 |
Scores use a 0–100 scale; higher is better. Source: OpenAI’s October system card, Tables 5 and 6. These are model evaluations, not an Intelligent UI usability score. The mixed results matter: improvement on one benchmark does not imply improvement on all of them. HealthBench numbers are length-adjusted, and the card cautions that challenging safety tests do not represent ordinary production failure rates.
For an interface feature, we would also want independent measurements of calculation accuracy, working controls, accessibility, mobile usability, and how often users complete their task without repairing the generated experience. The sources reviewed do not supply a single independently validated Intelligent UI benchmark covering those questions.
OpenAI reports stronger jailbreak resistance and factuality on most tested measures, while also disclosing regressions in some safety evaluations. The detailed system card is more informative than a blanket claim that every behavior has improved.
How to try it
- Open ChatGPT on the web or an updated supported mobile app, and use the Chat tab.
- Look for GPT-6 in the model menu once the rollout reaches your account.
- Choose a supported effort level. Pro reasoning is not the Intelligent UI option.
- Ask for a task where interaction helps. No special activation phrase is required.
- Test the controls and check the answer’s sources, assumptions and calculations.
If it is missing, account rollout, app support and workspace permissions are the first things to check. On the web, Settings → Layout and visuals can reduce visual elements, although it does not guarantee a completely text-only experience. Source: OpenAI’s usage guidance.
What this changes
Our view is that Intelligent UI should be judged by whether it reduces work for the person asking the question. A useful generated interface makes a relationship visible, exposes assumptions and helps you explore alternatives. A weak one adds decoration or hides a fragile calculation behind convincing controls.
Try it with a small task you already understand. Change an input, inspect what changes, and ask for an explanation of the result. That is a better first test than asking it to create an elaborate dashboard whose correctness is difficult to assess.
GPT-6’s wider rollout makes that experiment available to a much broader audience. The lasting significance will depend on whether these interfaces remain useful after the novelty wears off: clear enough to understand, reliable enough to check, and quick enough to improve the task at hand.
Editorial note: Access and pricing were checked on October 7, 2026. Official sources are linked beside the relevant claims. Benchmark figures are OpenAI-reported; the prompts and editorial assessments are Kingy.ai’s analysis.
The Kingy Brief
Get The Kingy Brief.
AI changes, original tests and one practical thing to try. Fridays at 09:00 Vancouver time.
Free · Double opt-in · Unsubscribe anytime
Signup help and newsletter schedule
Signup form provided by Beehiiv. After submitting, check your inbox for "Confirm your subscription to The Kingy Brief" and open its confirmation link. Check Spam or Promotions if you cannot find it.
Fridays at 09:00 Vancouver time: source-checked AI changes, original tests and one practical thing to try. The weekly restart begins October 9, 2026. We skip a week when there is not enough verified material. Free. Unsubscribe anytime.