The fact that it was already busy the last few days for The new model releasesOpenai strengthens a week Preliminary viewing of studies GPT-4.5The company advertises the new system as its largest and best chat model. In early testing, Openai says that people believe that the GPT-4.5 is a natural spoken one, with the ability to transmit heat and demonstrate a kind of emotional intelligence.
In one example, which is separated by Openai, a person says to Chatgpt that he is going through difficult times after the failure of the dough. Where are previous models of the company, including GPT-4O And O3-MiniThere may be a condolences with a person before offering a long list of undesirable advice, GPT-4.5 accepts another beat. “Do you want to talk about what happened, or do you just need to be distracted? I am here in any case, ”says the chatbot when power from GPT-4.5.
The profit shown by the GPT-4.5 are the result of achievements made in uncontrolled training. With uncontrolled learning, the machine learning algorithm is provided with a non -stamped data set and leave on their own devices to find templates and ideas. GPT-4.5 does not “think” as modern models of the company's reasoning, but when training the new model Openai made architectural improvements and provided it with access to more data and calculate the power. “The result is a model that has a wider knowledge and a deeper understanding of the world, which leads to a decrease in hallucinations,” the company says.
Speaking about reduced hallucinations, Openai measured how much better the GPT-4.5 in this regard. When carried out SimpleqaThe Openai software standard, which checks the large language models to their ability to respond with “simple but complex issues of knowledge”, GPT-4.5 knocked out O3-Mini, GPT-4O and even O1 with a hallucinations level of 37.1 percent. Obviously, the new model does not solve the problem of AI hallucinations at all, but this is a step in the right direction.
Despite its relative strengths compared to the GPT-4O and O3-Mini, GPT-4.5 is not a direct replacement of these models. Compared to the systems of reasoning Openai, GPT-4.5 is “a more general, initially smart model.” In addition, it is not initially multimodal as GPT-4O, which means that it does not work with functions such as Voice modeVideo or screening. It is also a “very large and intense model.”
It is best to think about GPT-4.5 as steps for Open Plans Plans systems in the future. In fact, Sam Altman said the same as at the beginning of this month when he shared the company's roadmapNoting that the GPT-4.5 will be “our last model, not chain thoughts”-licking the fact that the new system does not solve the problem, solving them step by step, as Openai's reasoning models do. Its successor, GPT-5, will probably integrate many of the latest Openai technologies, including it Frontier O3 ModelThe Openai field confirmed that today, saying that it plans to convey “unique strengths of the GPT-4.5, including wider knowledge, stronger intuition and greater“ equalizer ”,“ all users in future models ”.
At this time, ChatGPT for Subscribers may begin to use the GPT-4.5, starting from today, and PRO users and commands should access next week.