GPT-5 disgraced OpenAI during the launch

OpenAI carried out the activity that the world expected yesterday evening and the most advanced model GPT-5officially introduced. The model, which would give strengthening to Chatgpt, was coming, especially in particular. However, during the presentation about the GPT-5, there was a detail that attracted the attention of a number of users: Tables about GPT-5.

Many users shown about the performance of GPT-5 during live broadcast that the tables are problematicobserved. First of all, the tables show the impressive performance of the GPT-5, but when you examine it more closely, you can see that it is actually problematic.

The absence of the dimensions of the bar in the graphics reacted

For example, let’s look at the coding section in the mid -models deception ratio graph. When we look at the graph, we can see that the GPT-5 gets a smooth rate of 50 %. The O3 model compared is behind it with 47 %. What is strange here, sticks in the graph. So much O3’s bar looks bigger. In fact, it needs to be below because it performs more makus. So it can mix this way.

If we give an example, the table you see is also shown in the live broadcast. In this table, we can see that the scores of GPT-5 are lower than O3. However, the stick of the GPT-5 looks bigger and supposedly more smoothly misunderstandingWhy is it.

Apology from OpenAI

We fixed the chart in the blog guys, apologies for the unintentional chart crime 🙏

Can't wait for you to start using GPT-5 as we start rolling it out today!https://t.co/1x6yZ4k7TU pic.twitter.com/M1IchNbHwd

— pranav (@pranaveight) August 7, 2025

An OpenAI marketing employee said that they have corrected the problems in the tables by publishing an apology sharing about betting. This is in the blog post that the company shared about GPT-5 from its site problem solved. However, the fact that such a large company includes such flaws in the live broadcast, attracted the reaction of users.

There is no information about whether the tables used in the presentation are created with GPT-5. However, if this is the introduction of such a large model, this kind of error is not enough image. Moreover, considering that the model is much more sufficient in hallucination.