The successor of GPT-4 (and GPT 4.5, Code Name Orion, which is a transition model) will be launched in early August, according to US media The Verge, after some additional tests and a little delay. Many observers of the generative world of the arrival for the month of July.
Few details are known about this new model that will feed Chatgpt. “This is an experimental model that incorporates new research techniques that we will use in our future models,” said Sam Altman in X on July 19. “We believe he will love GPT-5,” added the OpenAi CEO.
A few days later, he announced other information about this future model in a podcast with the humorist Theo von. While trying the development model, he received a question by email that he did not understand. He sent it to GPT-5, who responded perfectly.
At that time, Sam Altman felt “unnecessary compared to AI”, believing that he should have answered the question, which GPT-5 had done instantly. “It was a strange feeling,” he said.
A unification and simplification of the offer
In addition to his new information, the head of OpenAI has already described GPT-5 as “a system that integrates a large part of our technology” last February. Many observers consider that in this new model it is an important step, including the unification of OpenAi models and technologies.
Therefore, it will include in particular the O3 reasoning model, which should no longer be available separately, according to a position on Sam Altman’s X of February 12, 2025. This strategic fusion is part of commercial efforts to simplify and combine the models it offers to achieve a more efficient system. This will prevent users from having to choose according to what they want to do: talk, reason or send multimodal tasks. “We want AI to give you satisfaction, that’s all. And we realized the complexity of our model and product offer,” Sam Altman said in the same publication.
The chain of thought in the heart of real reasoning?
GPT-5’s great novelty could be a true reasoning of the chain of thought. Therefore, it could administer a logic in several stages and better decision making. This would allow him to unify and strengthen his “reasoning” capabilities, taking into account multimodal entries and perform tasks with a single model.
I could also gain a lot of skills and overcome the most common coding problems, which should facilitate the lives of developers.
Finally, it could reduce its level of hallucination very strongly, in other words, it is more reliable in the answers it brings, even if it is always necessary to take with distance and caution the figures provided by tests such as simpleqa, which only partially reflect the reality and daily reliability of most of our interactions with chatpt.
GPT-5 will also be available in several versions, as Sam Altman had indicated. While free users will unlimited an unlimited way to a standard “level of intelligence”, chatgpt plus (23 euros per month) subscribers will have a more intelligent version. And this intelligence level will be “even higher” for the subscribers of Chatgpt Pro (229 euros per month).
Source: BFM TV
