OpenAI brings GPT-6 and Intelligent UI to ChatGPT's 1.2 billion weekly users
- OpenAI shipped GPT-6 inside ChatGPT on October 7, 2026, aimed at the 1.2 billion people who use ChatGPT each week, one month after the first GPT-6 models were limited to paid customers.
- The release adds Intelligent UI, which composes answers from graphics, tappable buttons, forms, charts and interactive experiences, choosing the format from the question: side by side for comparisons, an interactive diagram for explanations, plain text when that is the most useful response.
- A compiler renders the interface progressively as the model generates it, so elements appear without waiting for the full response to finish.
- Demo answers include an exploded 7-speed bicycle breakdown with five selectable systems (frame, wheels, drivetrain, brakes, cockpit) and a Sunday lamb roast plan with a lamb-and-potato scaling table for 4 to 10 guests.
Hacker News opinions
Is this OpenAI catching up with Anthropic's artifacts? And note the wording, they trained GPT-6 to compose responses with visuals, not a harness bolted on top.
They already had a simpler version of this, it was the obvious next step. What I want is tools wired in so I can pick a color or drag a slider by hand when precision matters.
It's strange that they're putting GPT-6 Sol into Chat instead of GPT-6.1 Sol, which is significantly more capable.
I've hit model not available limits on 6.1 Sol several times this week, on both my personal $200 Pro plan and my Codex enterprise seat. Switching to Astra works, but OpenAI looks capacity-constrained.
I really don't like having to open Work sessions for one-off questions just because they limit which models go into chat. The older models are too dumb for some things.
The gap between 6 Sol and 6.1 Sol was so short I doubt they share much. I suspect 6.1 is a heavier model rebranded as Sol after 6 turned out weak at agentic tasks.
6.1 is basically Astra minor. Two different models, shipped as Sol to recover from the GPT-6 launch mess.
6.1-sol landed seven days after 6-sol because they were bleeding to Anthropic's fable-5.1 and opus-5.5. Margins on it are thinner, so they have to push it somewhere to make the compute pay.
I'm building something similar for an internal project off Vercel's json-render, and it hasn't been hard, especially since faster models like 5.6 Luna showed up.
This looks bad to me. I want output that stays text so I can export and process it, and this will make an already resource-hungry app worse.
Visuals are just better for regular people. 'Help me cook X' or 'where do I stop for gas on the way to San Jose' isn't a text problem. If you want text only, ask for text only.
In the video they show interactive origami and bike assembly tutorials because people struggle with written manuals. I'd love an ELI5 tool for complex stuff, but I worry about outsourcing my own ability to do things to a mega corp.
This tracks with what I already see: ask Codex work mode something with a lot of nonlinear data and Astra or Sol will write a full HTML document with a short summary in chat.
I predicted last year there'd be a new UI protocol for agents, like HTML. Personalized UIs become the new browser interface, and browsers get replaced entirely.
Sounds horrible but plausible. They extracted all the value from open interoperable systems and now want closed proprietary ones.
So banks, utilities, government portals, medical records and taxes all go through chatbot APIs? Wikipedia, Netflix and the NY Times just hand over their content? That isn't happening.
The intro video is lame, those things never happen in real life.
Do you never assemble a bike or paint a room? For people with ADHD or kids who learn by watching, the video is spot on. Text alone is a bad UI for a lot of tasks.
Why not just stream HTML instead of a compiler? Presumably because an element can't render until it's closed.
Written manuals come with pictures too, so the ad is dishonest. Though the real target is Anthropic's text-only answers, not paper pamphlets. I get the best explanations by going a few sentences at a time back and forth, so I hope this UI isn't so aggressive I can't prompt it away.