If you are a chatgpt user, your new button may have caught your attention: Reason. But what does Razonar in Chatgpt mean? In the following sections I tell you everything about this function. Not only do I explain what exactly does and what language model uses, but I talk about the background and The reasons that have led Openai to introduce this functionnot only for paid users, but for everyone. Don't miss it!
Reasoning in chatgpt is a function that is Available for everyone. The only exception is that you are using this service without an account. In that case, you will not be able to activate the reasoning model. Now, whether you pay for chatgpt plus or pro as if you don't, Reason in chatgpt It is a characteristic available for you.
And how is it activated? Well, before sending a request so that the AI resolved, it is enough to press on the button Reasonas seen in the lower image:
This will make it marked and turn blue. This button, represented with a bulb icon, makes Chatgpt solve the request with the OpenAI GPT-O3 MINI Reasoning Model. But how does it work exactly?
Reasoning models They also generate text from text. In that sense, they are identical to traditional language models. That is, the function of both is the same: give answers to the user. What happens is that, with some complex tasks, especially Those that include mathematical or logic problemslanguage models usually invent the answer.
They will only offer a correct answer In the event that the riddle or problem is part of your training. If it is a totally new or invented enigma from scratch, then normal is not able to solve it. When I talk about traditional language models, I mean models such as GPT-4O or Deepseek-V3. And how is this problem solved?
Openai states that, if more computing time is dedicated to the question and the model is forced to imitate human reasoning, it is possible to obtain correct and free responses of hallucinations. Then, GPT-O1 launched, a reasoning model that “thinks” for a longer time. What does really is Desgrate the user's request and apply more complex connections between the different concepts. Then, solve the request step by step to provide a correct answer.
Taking all this into account, so we can summarize what happens when we activate the option of reasoning in Chatgpt:
- The GPT-O3 mini model is activated that simulates human reasoning.
- The model dedicates more resources to elaborate an answer to increase the possibilities of being truthful.
- After following a reasoning chain and solving the step step by step, use the language model to elaborate the text where the answer is included.
Reason in chatgpt in response to Depseek
O1 and O1 MINI reasoning models are only available for chatgpt payment users. However, since the arrival of Deepseek-R1, a more efficient and free reasoning model, OpenAi has responded by adding this capacity for all users. Mainly, this is the reason why you can now use the function of reasoning in chatgpt freely. Check out the Deepseek and Chatgpt comparison that we have published.
Testing the function of reasoning in chatgpt
So that you understand what supposes better Activate the Razonar Function in ChatgptI have done some tests and I have documented them. In them, I have compared Chatgpt's responses with the reasoning function deactivated with which I have obtained with the reasoning model. You will also see some Deepseek response, the Chinese AI that also has a model of these characteristics, Deepseel-R1.
Calculating the letter of the DNI
The first thing I'm going to ask Chatgpt is that Calculate the lyrics of a DNI that I am going to invent. In case you did not know, the Spanish ID has an associated letter that is calculated with a algorithm that is public. This is the request:
- Calculate the lyrics of a Spanish ID with the number 14515488
First of all, you should know that the correct answer is G. That is, the complete ID would be 14515488g.
Without activating the option to reason in chatgpt, this is the result I get:
On the contrary, if you active the chatgpt reasoning option, this is the result obtained:
As you can see, after “thinking” for a few additional seconds, the reasoning model is able to correctly solve the problem, while the language model without this additional layer returns a wrong response.
It should be clarified that, depending on the ID you enter, it is possible that the reasoning model fails. The only thing that changes is that There are less possibilities that it is wrong, but it is not impossible either.
Solving a little known enigma
Using well -known enigmas with language models ensures a correct answer. This is because That same response was surely included in training. But what if we use a little known riddle? That's where the Reasoning models They should be able to offer a correct answer, even without prior training.
After See Xavi Mitjana's video where compares the OpenAi O3-mini reasoning model with that of Deepseek, I have taken this Prompt To help you see The differences between a model like GPT-4O and O3 Mini. This is the problem:
- If a mathematics teacher affirms to his students that “in the world there are three kinds of people: those who know how
Obviously, this statement is false. But the grace of this riddle is that the teacher was wrong to count the groups of people who exist and that, in this way, It was placed in the group of those who do not know how to add. Chatgpt without reasoning is not able to see the joke of this riddle:
Meanwhile, after just a few seconds reasoning, Chatgpt with the O3 Mini model yes manages to realize this detail:
My advice is that you test chatgpt activating and deactivating this function To see the differences in the answers offered by both models.
And you, what do you think of the function to reason in Chatgpt? Do you think you have practical applications in our day to day? Leave me a comment below. We read!
