Openai is Starting GPT-4.5 todayHis latest and largest AI language model. GPT-4.5 will be available as a research preview for CHATGPT PRO users. Opeli calls the edition of “the most competent model”, but he warned initially that GPT-4.5 is not a border model and may not work as well as O1 or O3-Mini.
GPT-4.5 will have better writing opportunities, improved world knowledge and what OpenAi calls “a sophisticated personality over previous models.” Opeli claims that the interaction with GPT 4.5 will be more “natural”, adding that the model has better recognized patterns and drawing of connections, making it ideal for writing, programming and “solving practical problems”.
However, he notices OpenAi that he will not introduce enough up-to-date possibilities to be considered a border model. “GPT-4.5 is not a border model, but it is the largest OPENAI LLM, it improves GPT-4 computing efficiency by over 10x,” Openai said in the document It leaked before the announcement. “It does not introduce 7 net limit capabilities compared to previous editions of reasoning, and its performance is below the results of O1, O3-Mini and deep research on the most important ratings of readiness.” Since then, Opeli has removed this mention Updated version document.
It was previously reported that Opennai used his modeling model O1, related strawberries, to GPT-4.5 train with synthetic data. Opeli claims that it has trained GPT-4.5 “using new supervision techniques in combination with traditional methods, such as supervised tuning (SFT) and learning to strengthen based on human feedback (RLHF), similar to those used in GPT-4O.”
Despite some of its restrictions, GPT-4.5 hallucinate much less than GPT-4O, according to OPENAI and slightly less than its model O1. “We adapted GPT-4.5 to be a better collaborator, thanks to which the conversations seem warmer, more intuitive and emotionally refined,” said Raphael Gontijo Lopes, a researcher from OpenAI, during live broadcast. “To measure this, we asked human testers to evaluate it against GPT-4O, and GPT-4.5 basically exceeded every category.”
In the post on xThe CEO of Opeli, Altman himself, admitted that GPT-4.5 is a “gigantic, expensive model” and that “will not crush comparative tests.”
After running Pro users, OPENAI claims that GPT-4.5 will move into Plus and team users next week, and then to Enterprise and Edu users. It is too Available now in the Azure AI of Microsoft The platform, along with up-to-date models made of stability, Korem and Microsoft.
We revealed ourselves Notebook Last week, that Opeli planned to launch GPT-4.5 until the end of February, and GPT-5 at the end of May. The president of Opeli, Altman himself, called GPT-5 as “a system integrating many of our technologies” and will contain a up-to-date model of O3 OpenAI reasoning, which the company teased during 12 days of Christmas advertisements in December.
While Openai released O3-Mini last month, OPENAI only sends O3 as part of the upcoming GPT-5 system. This is in line with OpenAI’s goal to combine your huge language models to finally create a more talented model that can be called artificial general intelligence or agi.
Update, February 27: He noticed that Opeli changed the GPT-4.5 system card and added information from Altman itself.