OpenAI launches GPT-6 to 1.2 billion users with on-the-fly intelligent UI generation and streaming reasoning

OpenAI launches GPT-6 to 1.2 billion ChatGPT users

OpenAI has rolled out GPT-6 to its full 1.2 billion ChatGPT users, shipping two capability changes that Aarush Selvan, a ChatGPT product manager, describes as meaningfully different from prior model upgrades rather than incremental improvements.

“Today, we're launching GPT-6 and ChatGPT for all of our 1.2 billion users. ChatGPT can now control not just what to say, but how to show its answers and can control its interface. ChatGPT can now answer while thinking — it can start answering while it's still continuing to work on the problem.”

Intelligent UI

The bigger architectural shift is what OpenAI is calling Intelligent UI. Previously, ChatGPT responses lived inside a fixed set of pre-built cards and markdown templates. The model could populate a weather widget or render a chart, but the interface itself was static and designed in advance. With GPT-6, the model generates the UI on the fly for each query, deciding both what to say and how to present it.

Selvan gives a concrete example: ask ChatGPT to explain the offside rule in soccer, and instead of producing a text walkthrough, the model might generate an interactive diagram you can manipulate directly. Ask for a recipe, and it might add a copy button to the shopping list, not because that button was pre-built, but because the model determines that's the most useful thing to include.

OpenAI built a proprietary in-house component library to make this work at scale. The constraint with earlier visual generation, as Selvan describes it, was compute and latency. The new library lets the model generate UI in a token-efficient, streamable way, fast enough to feel like a normal chat response, cross-platform, and cheap enough to offer to all users without usage caps or additional charges.

Streaming reasoning

The second change addresses latency on complex queries. Previous reasoning models required the model to finish thinking before it could start responding. GPT-6 now begins answering while still working through the problem, which Selvan describes as a subtle but meaningful change in how users experience ChatGPT on harder questions.

What it means for product and design

Selvan is candid that Intelligent UI is forcing OpenAI to rethink how it builds products internally. Designers can no longer mock up a specific card and specify when it will appear, because the model generates the interface dynamically. OpenAI has built new design tools and a development stack that puts design teams working directly alongside researchers to train better design outputs into the model itself, rather than pre-specifying layouts.

On the product management side, Selvan argues the shift to nondeterministic UI compounds an already significant transition. The field was already adapting to nondeterministic text. Now the interface itself is variable, which changes how OpenAI measures progress, runs evals, and collects user feedback.

Longer-term direction

Selvan frames chat as evolving toward an operating system that updates its own interface depending on what a user is trying to accomplish. The near-term practical version is what launched today. The implied direction is a surface that generates whatever UI a task requires rather than constraining users to formats OpenAI built in advance.