OpenAI announced on 7 October that it is bringing a new GPT-6 model to ChatGPT with Intelligent UI, a capability the company says can combine written answers with visuals and controls that work inside a conversation. OpenAI said the first GPT-6 models reached paid customers in September and described the new model as intended for ChatGPT’s more than 1.2 billion weekly users.
Key points
- OpenAI says GPT-6 can put graphics, buttons, forms and charts directly into ChatGPT responses.
- TechCrunch reports a global rollout for Pro, Plus, Business and Enterprise on 7 October, with Free and Go scheduled for 8 October.
- OpenAI claims GPT-6 Instant starts answering web-search questions 44% sooner on average than GPT-5.6 Instant.
GPT-6 puts controls inside answers
OpenAI says it trained GPT-6 to choose among text, pictures and interactive elements according to the question. A response might place a comparison side by side, offer a diagram whose parts can be explored, or remain plain text when that fits the request. The change concerns the answer itself: a button or form can be used in the conversation, rather than serving as an illustration alongside a block of prose.
Planning a Sunday roast could become easier to follow if the recipe appeared alongside its timing, as OpenAI’s example proposes. The useful distinction would be between reading instructions and working through a plan whose parts are visible together; whether the generated plan is sound would still depend on the answer GPT-6 produces.
In a demonstration for journalists, OpenAI generated diagrams in response to a question about how an aeroplane wing produces lift, TechCrunch reported. The company’s product manager Aarush Selvan described ChatGPT as having “predominantly been a text-based interface”. That demonstration showed a visual explanation produced for a particular prompt; OpenAI’s broader proposition is that the model can decide when such an explanation is useful across different questions.
The controls need not dominate every exchange. Users will be able to reduce the number of visuals in their responses, TechCrunch reported. OpenAI also gives a plain-text answer as one possible choice for GPT-6, leaving the format to depend on what was asked rather than making every reply into a miniature application.
OpenAI’s compiler assembles responses as they arrive
OpenAI built a library of components that can be streamed and a compiler that processes an interface while the model generates it, Unite.AI reported. Think of the components as the prepared pieces of a page — a chart, a control, a space for text — and the compiler as the part that puts the model’s choices on screen. According to Unite.AI, this arrangement lets an interface appear progressively instead of waiting for the whole response to be finished.
OpenAI says it expanded training so GPT-6 can make decisions about layout, visuals and interaction, and assessed the resulting interfaces for clarity, usefulness and completeness. The company also says it has more work to do on the model’s design judgement and on the range of interfaces it can create. A usable control therefore depends on more than drawing it: the model has to choose an appropriate element and place it where the answer can be followed.
For developers building around ChatGPT, OpenAI’s account places interface-making within a model response. AI Affairs previously covered sidebar homes and event-driven automations for ChatGPT plugins.
GPT-6 Instant starts web answers sooner
OpenAI says GPT-6 can begin answering while it continues to think, adding to a response in stages. On the company’s evaluation of questions requiring web search, GPT-6 Instant started answering 44% sooner on average than GPT-5.6 Instant, Unite.AI reported. The measured event was the start of an answer, not the completion of one, and the figure comes from OpenAI’s evaluation.
ChatGPT’s 7 October access differs by tier
The timing of access has been described differently. GPT-6 is now available to all users, Breaking the News reported. The global rollout on 7 October covers Pro, Plus, Business and Enterprise users, while Free and Go users are scheduled for the following day.
