OpenAI’s GPT-5 rollout is not going smoothly

The performance of the 4O model was better for me on at least one of these mathematics problems. Unfortunately, Openai slowly neglects these old models-including the previous GPT-4O GPT-4O and the powerful O3 thinking model – For Chatgpt users, although they will continue to be available in the API programming interface for developers for the expected future.

Not good in coding as the criteria indicate

Although Openai’s internal standards and some external standards for an external party have shown GPT-5 to excel all other models in coding, It appears to be in the real world, it seems that Claude Obus 4.1, which has been recently updated, is doing a better job in certain tasksThat is, completing the application required for the user or creating programs according to his specifications. See an example below from Justin Sun posted on X:

In addition, a report issued by the SPLX security company found that the Openai’s inner safety class left major gaps in areas such as alignment of business and weakness to demand injection and logical attacks.

While the anecdotal, it appears that checking the temperature on how the model is attached to the adopters of artificial intelligence early indicates a cold reception.

The former artificial intelligence influencer and Gogler Bilawal Sidhu published an opinion poll On X he asks for “examination of the atmosphere” from its followers and the enthusiasm of the broader user, and so far, with 172 votes, The overwhelming response is “Kinda Mid”.

As he wrote AI leaks and news account, books, “The overwhelming consensus on the GPT-5 of both X and Reddit Ama is significantly negative.”

Tibor Blaho, a main engineer in AIPRM, the famous AI poster and a news poster on X, many problems with ChatGPT-5 in an excellent post, highlighting that one of the new Marqueee features -“automatic router” in ChatGPT chooses to think or not think of the primary GPT-5 model depending on the difficulty of querying-becoming one of the main complaints, Given that the model appears to be default for a non -thinking mode for many users.

Competition awaits the wings

Thus, and Feelings about Chatgpt-5 are far from positive globally, which highlights a big problem in Openai Since it faces an increasing competition from the main competitors of the United States such as Google and Noteropic, and a growing list of free, open and powerful LLMS, the source offered by many US models.

Take Alibaba Qwen team from artificial intelligence researchers, Only today updated high -performance QWEN 3 model for a symbolic context of one million people – Give users the ability to exchange approximately 4X information with the model in one back and forth reaction such as GPT-5 offers.

Looking at the other large version of Openai this week-which receives the new GPT -SS models, the new source-mixed reception from the first users, do not search for things for the first company designated by users at the present time (700 million weekly users active from ChatGPT as of this month).

In fact, this is also evident by Polymarket users with an overwhelming majority after the GPT-5 version. Google may have the best model for Amnesty International by the end of this month, August 2025.

Other power users like others Amnesty International and Matt Schumer, who have received early access to GPT-5 and positively around it in a review here, OPINED that views will change with the discovery of more people the best ways to use the new model and adjust their integration methods:

Although it is still the first days for GPT-5-feelings can change significantly with more users with their hands and trying them at various tasks- Early indicators do not seem like this version of “home run” for Openai In the same way as previous versions such as GPT-4, or even the latest 4O and O3. This is a disturbing indication of a company that just collected another financing round, but it is still unanimous due to the high research and development costs.