OpenAI is launching GPT-4.5 today, its newest and largest model. GPT-4.5 will be available as a research preview, and OpenAI is calling it the company’s “largest and most knowledgeable model yet,” but warning that it’s not a frontier model and might not perform as well as o1 or o3-mini.
GPT-4.5 will have better writing capabilities, improved world knowledge, and what OpenAI calls a “refined personality over previous models,” but it won’t introduce enough new capabilities to be considered a frontier model. “GPT-4.5 is not a frontier model, but it is OpenAI’s largest LLM, improving on GPT-4’s computational efficiency by more than 10x,” OpenAI says in a document that leaked ahead of its announcement. “It does not introduce 7 net-new frontier capabilities compared to previous reasoning releases, and its performance is below that of o1, o3-mini, and deep research on most preparedness evaluations.”
It was previously reported that OpenAI was using its o1 reasoning model, code named Strawberry, to train GPT-4.5 with synthetic data. OpenAI says it has trained GPT-4.5 “using new supervision techniques combined with traditional methods like supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), similar to those used for GPT-4o.”
Despite some of its limitations, GPT-4.5 hallucinates a lot less than GPT-4o, according to OpenAI, and slightly less than its o1 model. Leaks earlier this week suggested GPT-4.5 will be limited to ChatGPT Pro users, much like how Operator was released. OpenAI is about to hold a stream where it will share more information on how GPT-4.5 will be released.
I revealed in Notepad last week that OpenAI was planning to launch GPT-4.5 by the end of February, and GPT-5 as soon as late May. OpenAI CEO Sam Altman has referred to GPT-5 as a “system that integrates a lot of our technology,” and it will include OpenAI’s new o3 reasoning model, which the company teased during its 12 days of Christmas announcements in December.
While OpenAI released o3-mini last month, OpenAI is only shipping o3 as part of its upcoming GPT-5 system. This aligns with OpenAI’s goal to combine its large language models to eventually create a more capable model that could be labeled as artificial general intelligence, or AGI.
