OpenAI put GPT-6 into ChatGPT for everyone, with an interface the model builds
For three years a chatbot answer has been a paragraph. On 7 October OpenAI started shipping answers that are small interfaces the model builds for the question, and put them in front of free users the next day.
OpenAI brought GPT-6 into ChatGPT with Intelligent UI: replies that combine text, visuals and interactive controls assembled for the particular question rather than returned as prose. Plus, Pro, Business and Enterprise get GPT-6 Sol in the Chat tab from launch day. Free and Go get GPT-6 Luna from the day after. The Pro reasoning option continues to run GPT-6 Astra, which does not support the new interface. This is the first time the GPT-6 family has reached the free tier at all, and it comes nine days after OpenAI halved the price of that family.
Why this one is different
Every assistant has been converging on the same shape: a text box, and prose coming back. Intelligent UI breaks that by letting the model decide what the answer should look like, which moves the product from a conversation to a generated application. The second thing is the distribution. Free users have been a tier behind for two years; they are now one day behind, on the cheap model of the current family. That is a decision about cost, and it was only possible after the price cuts.
The model no longer writes the answer. It builds it.
How we got here
- 22 Sep 2026GPT-6 Sol and Luna arrive at half the price of the GPT-5.6 tier.
- 29 Sep 2026DevDay ships dots, GPT-6.1 Sol and a $500 tier.
- 7 Oct 2026GPT-6 with Intelligent UI reaches ChatGPT, and the free tier a day later.
What it does and does not mean
Nothing has been published about what this costs to serve, which is the number that decides whether free users keep it. An answer that builds an interface spends more tokens than one that writes a paragraph, and OpenAI has just spent a fortnight cutting prices rather than raising them. There is also no accuracy claim attached: a generated interface can be wrong in ways prose is not, because a control that does the wrong thing looks exactly like one that works. What is clear is the direction of the competition. Anthropic spent the fortnight making the model cheaper; OpenAI spent it making the answer look different, and both are chasing the same volume.