OpenAI’s Rollout of GPT-5 Is Going Terribly

On Thursday, Openai officially revealed GPT-5 in the world. The very mediated presentation was sparse on many specific benchmarks comparing GPT-5 to its previous models, but Openai staff were categorical: this model is the best, the most competent and the most powerful to date.

GPT-5 has its enemies

Many users who tested the GPT-5 driving in the 24 hours, which followed, however, in disagreement. A visit to R / Chatgpt is sufficient to see the scope of the situation: the first page is full of messages complaining of the current state of the model, in particular: “GPT-5 is the biggest [piece] Garbage even as a paid user, “” Openai has just drawn the biggest bait and switch in the history of the AI and I have finished “and” the deployment of Chatgpt-5 is an unmountained disaster “.

One of the most important complaints concerns OpenAi’s decision to depreciate the previous models, which the company announced without ceremony during the GPT-5 presentation. GPT-4O, O3, 4.5 and other models are no longer available to use. In the future, users will only have access to GPT-5 and its subsequent models (for example GPT-5 Mini). Many users are upset that Optai has won the previous models overnight without warning, especially when they believe that replacement does not offer the same experience. Some have even canceled their subscriptions accordingly.

I know that people use Chatppt for therapy, and I am aware that people have trained deep attachments to technology, but I admit that I was a little shocked to read some of the emotional reactions to lose access to these models. In an article, a user detailed how he was based on individual models for different tasks: he would use 4o for creative ideas, O3 for logical problems, O3-Pro for in-depth research and 4.5 for writing tasks. Another user talked about how he used 4o to help his anxiety and depression, because, in his opinion, the model felt “human”. They believe that people cry the loss of 4o, who follows, at least with other specific 4o positions. There are people who Really like these models, and are splendid After their withdrawal.

But beyond mourning, some users think that GPT-5 is not very good. If you ask the model how many times the letter “B” occurs in the word “blueberry”, he would have said “three”: once at the beginning, once in the word “blue” and once in “Berry”. This is not necessarily a new problem – the LLMS also had trouble spelling “the strawberry” – but it is not an excellent look for the “best” model of Openai of all time. An X user has highlighted an example of GPT-5’s incapacity to resolve a “simple linear equation”, compared to the ability of Google Gemini 2.5 to solve it without problem, while this user has displayed the generation of GPT-5 of a map of the United States, with most states labeled with fine party.

Some users have teased Openai on its waves comparative analysis data. Rhys on X Sarcastically published “These GPT-5 numbers are crazy”, and joined a graphic that has traced each GPT version by number (GPT-1 lands at “1” on the Y axis, GPT-2 to “2”, and so on until you reach GPT-5 to “5.”.

This tweet is currently not available. It can be loaded or has been deleted.

There are also criticisms of automatic switching, one of the main characteristics of GPT-5. Free and more Chatgpt users are unable to choose the specific model, but at the sight of Openai, this is a good thing. The GPT-5 is supposed to be smart enough to choose the right model for you according to your request: simple questions use lower models, while more complex requests use the most powerful models. But if Openai is so sure that it is a good thing, why does he always offer the possibility of manually changing models, as long as you pay $ 200 per month for a professional plan?

Not everyone agrees that GPT-5 is bad, be careful. There are users who seem to appreciate the model, appreciating concise responses and fast performance. But the majority of the speech I see on social networks and forums are neutral to negative. Even messages It seems first positive to end up criticizing the model:

This tweet is currently not available. It can be loaded or has been deleted.

4o lives, for the moment

Since the start of this play, Openai has responded to the backlash. CEO Sam Altman has published a series of updates on X which seem to go back to the decisions that users have criticized: the rate of rate will double for Chatgpt users more for the moment; The GPT-5 should seem more intelligent from today; It will be easy to see which model responds to a given request; And the manually choice of the reflection model will be simpler. Altman has also recognized that the initial deployment will be slower than expected, which is logical because I still do not have access to the new model.

But the group’s biggest announcement should be a new welcome for many users: 4O is back, at least for users more. If you pay $ 20 per month for chatgpt, you can continue to use 4O for the moment. Altman says that the company looks at the use and will make a decision on the duration for the models inherited in the future.

I am curious to know how users react in the future: those who canceled success to continue using 4O? Again, why bother, if Openai plans to withdraw this model again in the future? One thing is certain: this is probably not the way Openai expected that the deployment of GPT-5 takes place.

Disclosure: Ziff Davis, the parent company of Lifehacker, in April, filed a complaint against Openai, alleging that she broke Ziff Davis Copyrights in the training and exploitation of her AI systems.