ChatGPT's Intelligent UI Skips the Pro Thinking Level and Older Desktop Apps, and Publishes No Usage Figures
A.I. / news
ChatGPT's Intelligent UI Skips the Pro Thinking Level and Older Desktop Apps, and Publishes No Usage Figures
OpenAI put GPT-6 into the Chat tab on October 7 with generated buttons, forms and charts. The announcement gives one speed number and no accuracy figure.
OpenAI began rolling GPT-6 and a feature it calls Intelligent UI out to ChatGPT's Chat tab on October 7, letting the model answer with buttons, forms, charts and small tools instead of text alone. Plus, Pro, Business and Enterprise users got it first, and Free and Go users the following day, according to Search Engine Journal.
OpenAI's own announcement page returned an access error when this report tried to fetch it, so the details below come from two outlets that quote it.
What shipped on October 7
GPT-6 decides each response's layout from the question. A reply can be plain text, or it can include tappable buttons, forms and charts. Users can also ask for a tool, such as a calculator for savings growth or a bill splitter.
Gizmodo, in a story by AJ Dellinger, says the model builds these from a component library it was trained on. OpenAI's examples included a 7-speed bike breakdown with a button for each part, and a lamb roast plan whose button rescales ingredients for the guest count.
Which models and which tiers
| Tier | Model in Chat | Access |
|---|---|---|
| Plus, Business, Enterprise | GPT-6 Sol | October 7 |
| Pro | GPT-6 Sol, with Astra at the Pro thinking level | October 7 |
| Free and Go | GPT-6 Luna | October 8 |
Enterprise access depends on a workspace administrator. GPT-6 replaces the "Latest" option in the model menu, and GPT-5.6 Sol stays as a separate choice, per Search Engine Journal.
Sol and Luna first launched on September 22 for ChatGPT Work, Codex and the API. Those products' models are unchanged by this release.
Where it does not work
The limits are specific. Intelligent UI is not supported at the Pro thinking level, which runs on GPT-6 Astra. It is not supported in the older ChatGPT desktop apps for macOS and Windows.
It is rolling out gradually on the web and in updated apps, at thinking levels from Instant to Extra High. On the web, a "Layout and visuals" setting reduces the visuals, though some may still appear.
Search Engine Journal notes that the announcement does not say how sources will appear inside charts, comparisons or generated tools.
The one number OpenAI gave
For questions that need web search, GPT-6 Instant starts answering about 44% sooner on average than GPT-5.6 Instant. That is OpenAI's figure, and no baseline latency in seconds came with it.
- GPT-5.6 Instant100 index
- GPT-6 Instant (about 44% sooner)56 index
Source: Search Engine Journal, citing OpenAI, accessed 2026-10-09; vendor-supplied
On harder problems, OpenAI says only that GPT-6 addressed the key aspect of the question more often than GPT-5.6 in an internal test. No percentage was given.
There are no adoption, usage or error-rate figures for the interface itself. Gizmodo lists two risks: faster token use, and a new layer of hallucination risk when a wrong number sits inside a chart.
Context
Google introduced a similar Generative Interface with Gemini 3 in 2025, according to Gizmodo, which says it did not clearly help Google pull ahead. ChatGPT already had interactive charts from June and interactive modules for more than 70 math and science topics from March.
Gizmodo also notes OpenAI announced image-based ads in ChatGPT responses earlier the same week.
For how Luna is priced against rivals, see Claude Haiku 5.5 at the same list price. For another generated-interface product aimed at consumers, see Google Playground.
Neither outlet reports a date for Pro-level support or for the older desktop apps.
Sources
More in A.I.
- 01OpenAI Fires Three Safety Staff Nine Days After Publishing Outside-Audit PrinciplesThe company confirmed the dismissals on Oct. 1 but has not said what information moved, who received it, or whether the three first raised concerns internally.
- 02Google's EmbeddingGemma 2 Adds Images, Video and Audio to a 740M EmbedderThe Apache 2.0 weights lift the code-retrieval score by 9.9 points but move the multilingual text score by only 0.21.
- 03Cloudflare's Clef Beats TypeSafe's Jev on Three Tests and Loses on OneThe Apache 2.0 decision model costs nearly six times as much per million tokens as Jev, and its smaller sibling trails badly on one intent benchmark.
- 04Qwen3.8-27B Has 6.78M Downloads and an Apache 2.0 Licence, but Its Card Gives Training Data One LineAlibaba's 27B dense model scores 61.7 on SWE-bench Pro by its own table. The card does not say what it was trained on or what hardware it needs.