GPT-5 has many versions, but Openai decides which one to use for you

Openai wants to simplify our lives, but maybe he has made a mistake in doing so. When you use GPT-5, you won’t know which GPT-5 version you will be using. Because although Openai has raised it as a “unified model”, it is actually using variant distress according to the scenario and, of course, according to what you pay. And not knowing what he is using at all times is a blessed? problem. Welcome to the era of the opacity of AI.

GPT-5 are many GPT-5. In his “system card“Openai stands out how GPT-5 It is a “unified system” that actually consists of several versions. Some more powerful, others less, and others that reason. But unlike what happened so far-you chose if you wanted to use the lighter model or one that would think more-GPT-5 will choose for you. As?

On the one hand we have the “external” names of the variants of GPT-5 and on the other the inmates. The “real -time router” will be responsible for deciding which is used at each moment. Source: latent.space

A “router” that chooses for us. To know what variant you have to use, GPT-5 uses a “real-time router that quickly decides which model to use based on the type of conversation, complexity, or the need for tools.” As they point out In Latent Space The idea is remarkable, and helps to avoid That name chaos We had so far. Of course: the only way to “force” the type of model to use is to express it clearly in the prompt when using chatgpt. Thus, if we ask him something and ask that “think a lot about this”, we will activate that reasoning model so that the router does not make mistakes by choosing perhaps a less capable model.

Opacity. Although OpenAi indicates that he uses these arguments to decide which GPT-5 variant USA, here we face the dichotomy of opacity. That GPT-5 chooses for us is on the one hand good and on the other bad. Simplifies the use of chatbot, no doubt, and that brings it closer to the novel public. But at the same time Advanced users steals control or those who have enough knowledge to know how to decide which version of GPT-5 would like to use. With an expected exception.

We can already say goodbye to this old model selector. For better, and for worse.

Beware of simplifying. The simplification problem of complex technologies is that it can cause unwanted side effects. For example, that users do not understand the differences between the models. That leads to reducing its ability to make informed decisions about which tool is more appropriate for each specific context. Openai raises something like a “technological infantilization”: do not worry, dear users, that GPT-5 will already do everything for you. Even choose how to work.

If you pay, choose. If you are one of those advanced users who know how Chatgpt Pro (200 dollars/month) In order to activate any of the different internal models of GPT-5

Correspondence between the “old models” and the “new models.” Source: OpenAi

How are they compared to previous models. The arrival of this new family of models assumes that now each of the above It has its replacement with one of the Variants of GPT-5. Thus, the old GPT-4O would be the new GPT-5 Main, while GPT-4O-mini will be GPT-5-Main-Mini and O3 will be GPT-5-Thinking. And so with the rest:

Goodbye, old GPTS. As explained in the version notes Published in OpenAI, the arrival of GPT-5 supposes goodbye to the models that were used so far in Chatgpt. Chatgpt Pro subscribers They can continue to have access To these ancient models, at least for now, but for other users the option is clear: the pre-GPT-5 models are retired, and among them:

• GPT-4O (will continue to be used for voice mode)
• GPT-4.1
• GPT-4.5
• GPT-4.1-mini
• O4-mini
• O4-mini-High
• O3
• O3-PRO

GPT-5 should explain your decisions. Although the idea of having a unified model is not bad, that lack of control and information about the way of functioning of GPT-5 is worrying. One of the solutions is to introduce that selection of ways (available in Chatgpt Pro), but another is even simpler: Chatgpt would be enough to indicate which version of the model has used at all times so that we know what to stick to … and if we can force that in such questions a more capable (or less) model is used.

In Xataka | The AI has so far a nice accessory: with GPT-5, OpenAi wants to place it in the center of all things