Two identical cars are parked side by side. Same bodywork, same colour, same badge on the boot. But under the bonnet of the first there is a small engine. Under that of the second, a big one.
That is roughly what is happening this week in ChatGPT.
GPT-6 for everyone
OpenAI announced on Wednesday that GPT-6, its new generation of models, was arriving in ChatGPT for all of its users, everywhere in the world. More than 1.2 billion people use it every week.
Paying subscribers have had access since Wednesday. Users of the free version and of the entry-level plan, since Thursday. All of them will see the same name: GPT-6.
Answers that become tools
The most visible novelty is called "Intelligent UI". Until now, ChatGPT mostly answered with text, sometimes long paragraphs. From now on, it can answer with charts, buttons, forms or small interactive tools, directly in the conversation.
In practice, a comparison can become a table, a budget a chart, a practical question a small tool you can handle. You no longer just read the answer: you use it.
OpenAI also announces that answers drawing on an internet search are starting to appear 44% earlier on average than with the previous version.
One name, two models
This is the detail that matters, and it is written in small print. For paying subscribers, "GPT-6" rests on GPT-6 Sol, OpenAI's mid-range model. For free users, it rests on GPT-6 Luna, its smallest model.
Both are tuned for everyday conversation, and for most ordinary questions, Luna will do very well. But on tasks that require long reasoning, a complex document or a difficult problem, the difference will be felt.
To gauge the gap, you only have to look at what OpenAI charges developers. GPT-6 Luna costs 0.10 dollar per million tokens on input and 0.50 on output. GPT-6 Sol costs 2 dollars and 10 dollars. Twenty times more expensive. That is the gap a single name covers in the app: the same badge on the boot, not the same engine.
The same trend everywhere
OpenAI is not alone in making this choice. That same week, Google reserved the free version of Gemini for its smallest model, as explained in our article on free Gemini. The difference lies mainly in how it is said: Google announces it clearly, OpenAI keeps the same name for everyone.
The logic is the same at both. The small models have become very good and very cheap, which makes it possible to offer them free to hundreds of millions of people. The large models remain for those who pay. In the United States, free users additionally see advertisements, which cover part of the bill.
What you need to know
If you use ChatGPT for free, you do get GPT-6, but in its lightest version. For the vast majority of everyday uses, that is enough, and it is often better than what you had before.
If a response disappoints you on a difficult task, do not conclude too quickly that "the AI can't do it". It may simply be the small engine reaching its limits.
If you are hesitating to pay, first test your actual usage for a few days. A subscription is only worth it if your tasks genuinely need the large model.
What we take away
GPT-6 for everyone, that is true. More visual and more practical responses are a real improvement.
But you have to read the label all the way to the end. In AI as in the automotive world, the same name can hide very different engines. Knowing which one you are driving saves a lot of disappointment.