Implementation GPT-5 OPENAI is not going smoothly

Launching a long expected recent OPENAI, GPT-5 model is for a rocky start At least.

Even Forgiving errors on the charts AND Voice data Yesterday Live presentation the recent model (actually 4 separate models and “thinking” mode that may be involved in three of them), and The variety of user reports has appeared since the GPT-5 release showing that it hurts badly When solving relatively easy problems, which are preceded by OpenAI models – and rivals from competitive AI laboratories – answer accurately.

For example, a data scientist Colin Fraser has published screenshots showing GPT-5 error with mathematical evidence in mind (or repeating 8,888 is after all 9).

It is too Algebra’s easy arithmetic was not succeeded problem This basic students could probably nail 5.9 = x + 5.11.

Using GPT-5 to judge their very own erroneous charts of OPENAI presentations also did not bring helpful or correct answers.

It also failed This tougher problem with the mathematical word below (which, truthfully, at the starting surprised this man …Although Elon Musk’s Grok 4 Ai replied accurately. To get a clue, think that flags in this case can’t be divided into smaller portions. They must remain in tact, like 80 separate units, so no halves or quarters).

Made older model 4O Better for me At least one of those mathematical problems. Unfortunately, OpenAI is slowly observing these older models-in this previous default GPT-4O and a powerful modeling model O3 – For CHATGPT users, although they’ll still be available in the application programming interface (API) for programmers in the foreseeable future.

Not nearly as good in coding as reference studies indicate

Although OpenAI internal references and some outer exterior have shown GPT-5 to surpass all other models in codingIN It seems that in real use, the recently updated Claude Opus 4.1 Anthropica seems higher to perform in the “one shooting” of some tasksThat is, supplementing the desired application or user software for its specification. See The following example from the developer Justin Sun published to X :

In addition, AREport from the SPLX security company It was found that the inner security layer of OPENAI left serious gaps in areas equivalent to adaptation of activities and susceptibility to a quick injection and darkened logical attacks.

Although anecdotal, checking the temperature of how the model copes with the first AI users seems to point a cool reception.

Ai influencer and former googler balance sidhu published a survey on X with a request for “climate control” from its observers and a wider base of users, and so far, with 172 votes, The overwhelming response is “a little half”.

And as Pseudonym leaks and information account writtenIN “The overwhelming consensus on GPT-5 with X and Reddit AMA is predominantly negative.”

Tibor Blaho, chief engineer in AIPRM and popular AI leaks and information poster on X, summed up many problems with Implementation of chatgpt-5 in a great postemphasizing that one of the recent features of the awning -Automatic “router” in chatgpt, which chooses the way of pondering or not pondering for the basic GPT-5 model depending on the difficulty of the inquiry-was one of the principal complaints, Given that the model seemed default, the default mode that did not think for many users.

A contest waiting for the wings

Therefore The sentiment towards ChatgPT-5 is far from universally positive, emphasizing a significant issue for OpenAI In the face of growing competition from the principal American rivals, equivalent to Google and Anthropic, in addition to the growing list of free, open source and powerful Chinese LLM offering functions that lack many American models.

Take ALIBABA QWEN team AI researchers, Who Just today he updated his highly efficient QWEN 3 model to have 1 million context tokens – – Giving users the opportunity to exchange almost 4x such a great amount of data with the model in one rear interaction/Forth, as GPT-5 offers.

Considering the other great edition of OPENENAI this week-none GPT-OPEN SOURCE models-they also have a mixed party from early users, they are not currently looking for a dedicated AI company by users (700 million lively ChatgPT users from this month).

Indeed, it is also illustrated by Market users of the Polimarket plants In the prevailing deciding on the release of GPT-5 to Google would probably have the best AI model by the end of this month, August 2025.

Other advanced users like Other AI co -founder and general director Matt Shumerwhich received early access to the GPT-5 i I blogged about it positively in the review hereIN He said that his views would change when more and more people invent the best ways to make use of the recent model and adapt their integration approaches:

Although this is still early days for GPT-5-a sentiment can change significantly, because more users get their hands and try various tasks- Early indications do not appear to be this is the “Home Run” release for OpenAI In the same way as earlier editions, equivalent to GPT-4, and even newer 4o and O3. And this is a disturbing indicator An organization that has just collected the next round of financingBut it stays unprofitable as a result of the high costs of research and development.