will launch a model that was not going to see the light

If you use Chatgpt frequently, most likely you are interacting with GPT-4O, the model that is activated by default. However, it is not the only one. When displaying the model selector, several alternatives appear: from specific versions of GPT-4O Designed for programming up to GPT-4.5, options centered on reasoning and other models in the test phase that are not directly available in the chatbot.

The problem is that this growing variety is generating some confusion: The nomenclature has become little intuitive And many users are not clear which model is the most appropriate for each task. Sam Altman, CEO of the artificial intelligence startup (AI), recognized the problem in mid -February and announced a road map focused on simplifying its product offer. Now we know that this plan has encountered obstacles.

The original promise. Although Openai presented the O3 family in December last year, only the O3 Mini variant, focused on offering better performance, ended up reaching the public. In its new road map, the company said that the full version of O3, capable of obtaining scores similar to those of human programmers in certain tests, would not be launched as an independent model.

According to the updated plan month and a half ago, GPT-5 would integrate part of the technology developed by the company, including the O3 reasoning capabilities. In theory, that would help order the models ecosystem. As part of that strategy, GPT-4.5 would mark a turning point: the latest model without “chain of thought”. That is, from there, all new OpenAI models would have integrated reasoning.

There come the changes. Part of what we have just seen in the previous lines will no longer be fulfilled. And it is not a filtration or a rumor: It has been Sam Altman himself Who has confirmed it in a message posted in X. “Change of plans: after all, we will probably launch O3 and O4-mini in a couple of weeks, and then we will do GPT-5 in a few months,” he said. In summary: O3, which was discarded as an independent model, you will see the light.

Click to see the original message in x

The executive has also dropped some interesting details. On the one hand, as can be seen from the previous message, both O3 and O4-mini will be available soon. In addition, he confirmed that GPT-5 will not take long to arrive, presumably this year. Of course, there are no concrete dates, and it remains to be seen if in that margin of time we will attend another unexpected turn, such as the one we are witnessing now.

What is behind the change of plans? The big question is what has happened in OpenAi to produce this new unexpected turn. It is not a radical change, but it arrives at a delicate time: the American company is building a solid reputation around its ads and its real ability to fulfill them. Such a movement can weaken that trust, as valued by both users and investors.

In relation to change, Altman explained that “there are many reasons for this, but the most exciting is that we will be able to make GPT-5 much better than we originally thought.” Far from staying in an optimistic statement, he also admitted that integrating everything without problems was “more difficult than we thought” and that they wanted to make sure everything tied, especially because they expect “an unprecedented demand.”

More challenges in Openai. It is no secret that Openai is subject to increasing pressure. The startup, which initially grew with the support of Microsoft, competes to maintain its leadership position in the development of AI and to ensure the necessary financing to sustain its ambitious rhythm. To all that is added a challenge not less: guarantee access to the computing capacity that your projects require.

Images | OpenAI | Screen capture

In Xataka | In Anthropic they have become obsessed with some trivial but key appearance for AI: that Claude will pass ‘Pokémon’