Openai introduced ChatGPT-5, calling it the fastest, smart and reliable AI in the ruler, with improved protection against hallucinations.
According to the Vectara platform, which is rating of error frequency models, the new AI really gives out less false facts than the GPT-4 (1.4%versus 1.8%), but only a little ahead of the GPT-4O (1.49%).
Interestingly, the ChatGPT-4.5 Preview showed an even lower indicator-1.2%, and the leader of 0.795%O3-Mini High Reasoning remained the leader.
Despite the improvements, the GPT-5 is still inferior to 4.5 accuracy, although it is significantly ahead of competitors like Gemini-2.5-Pro (2.6%) and GROK-4 (4.8%).
The launch caused a scandal due to the fact that Openai removed the old GPT-4, 4 and 4O-Mini models from PLUS subscriptions without warning. Many users complained that they lost their usual functionality, and some called it the “loss of an only friend.”
Sam Altman acknowledged the importance of the old versions for the audience and promised to temporarily return the GPT-4O for subscribers in order to evaluate the demand for “heritage”.
