OpenAI has released GPT-4.5, its most advanced and largest large language model (LLM) for artificial intelligence. The developer calls the release its “most knowledgeable model,” but warns that GPT-4.5 is not a breakthrough model and may not perform as well as o1 or o3-mini, which have reasoning capabilities.
Image source: OpenAI
GPT-4.5 offers improved writing skills, better knowledge of the world, and what OpenAI calls “improved personality compared to previous models.” The company claims that interactions with GPT-4.5 will be more “natural,” noting that the model is better at recognizing patterns and identifying relationships, making it ideal for writing, programming, and “real-world problem solving.”
OpenAI, however, cautioned that GPT-4.5 doesn’t introduce enough new features to be considered a cutting-edge model. “GPT-4.5 is not a breakthrough model, but it is OpenAI’s largest LLM, exceeding GPT-4’s computational efficiency by more than 10x,” OpenAI said in a paper leaked before the announcement. “It does not introduce seven new features compared to previous versions with reasoning capabilities, and its performance is lower than o1, o3-mini, and Deep Research on most benchmarks.”
OpenAI previously reported that it was using its o1 reasoning model to train GPT-4.5 on synthetic data. OpenAI itself said that it trained GPT-4.5 “using new supervision techniques combined with traditional methods such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), similar to those used for GPT-4o.”
«“We adapted GPT-4.5 to collaborate better, making conversations warmer, more intuitive, and more emotionally rich,” said Raphael Gontijo Lopes, a researcher at OpenAI. “To evaluate this, we asked human testers to compare it to GPT-4o, and GPT-4.5 came out ahead in almost every category.”
Despite some limitations, GPT-4.5 hallucinates significantly less than GPT-4o and slightly less than the o1 model, OpenAI said. It also demonstrates greater intuition and creativity, better understands what users mean, and “interprets subtle cues or implicit expectations with more nuance.”
GPT-4.5 is available today to users with a $200 per month ChartGPT Pro subscription, as well as researchers. The model is currently in preliminary research testing. The decision to release it in this form is driven by a desire to “better understand its strengths and limitations.” “We’re still exploring its capabilities and are excited to see people use it in ways we may not have anticipated,” OpenAI said. The company did not say when it would make the new model available to the wider public.
Last week, OpenAI reported that it plans to launch GPT-4.5 by the end of February, with GPT-5 following in late May. OpenAI CEO Sam Altman called GPT-5 “a system that brings together a lot of our technologies,” noting that it will include OpenAI’s o3 model. OpenAI released o3-mini last month, but the full o3 will only be released as part of GPT-5. The company is aiming to combine its large language models to create one more powerful system that can independently determine which resources to use to solve a given task.