OpenAI surprisingly introduced o1, a new model that promises an important advance in a metric that until now has not been focused enough to be the protagonist of this type of progress. It’s true, yes, there have been some rumors (quite a few, actually) circulating in recent days that the creators of ChatGPT are finalizing the details to introduce something new, but on the other hand, and given the frequency with which the company usually introduces new features, that’s a thing , which was almost readable based on previous experience.

After it was revealed in the first half of last year that OpenAI had decided to suspend the completion and launch of GPT-5, there were two big questions in the air. The first is, of course, when this self-imposed moratorium will end, and the second is what the company will offer during said dead end. For the first, we don’t have an answer at the moment, but for the second, we already had the first sample with the GPT-4o presentation last May.

So things are It could be expected that the next news would be the revision of some of its modelsperhaps a relevant update of GPT-4o, with the arrival of new improvements, an extension of its functions available to users of free ChtGPT accounts or something similar, that is, a revision of what already exists, not a new language model, even less one that proposes a relevant change in aspect , which is an unsolved problem in the world of artificial intelligence today.

However, that’s exactly it what OpenAI did with the presentation of o1, an artificial intelligence model that focuses on the ability to reason. The technology company defines this new model and its way of proceeding in a way that leaves no one indifferent, stating that o1 does not intend to stand out for the speed of the answers, but because it will “think” about them before offering them. users. This means that by emulating human behavior, it performs a more complete process of analyzing the challenge, which at least theoretically translates into more accurate answers and, depending on the questions, possibly more complex ones.

More interestingly, according to OpenAI, the learning capacity will be o1 as a result that the model is able to identify its own errors and after identifying them, take them into account so that they do not recur in the future. This novelty may not be particularly relevant for many users who do not need particularly accurate and correct answers, but it is necessary to implement it and use it in certain environments where reliability plays a key role, even more so nowadays. in which we see core services adopting these technologies.