OpenAI has unveiled its GPT-4.5 artificial intelligence language model. The company calls it its “most knowledgeable model yet,” though not its most powerful. The main advancement in GPT-4.5 is in improved user interaction.
“We aligned GPT-4.5 to be a better collaborator, making conversations feel warmer, more intuitive, and emotionally nuanced,” OpenAI researcher Rafael Gontijo López said during the presentation. “To measure this, we asked human testers to evaluate it against GPT-4o, and GPT-4.5 outperformed on basically every category.”
GPT-4.5 demonstrates improved writing skills, deeper knowledge of the world, and what OpenAI calls “enhanced personality” compared to previous models. The company says that interactions with the new model will be more natural and lifelike. In addition, GPT-4.5 is significantly less likely to generate false information compared to GPT-4o and o1.
OpenAI CEO Sam Altman said it was the first model that made him feel like he was talking to an intelligent and thoughtful person. At the same time, OpenAI admits that GPT-4.5 does not offer enough new features to be considered a leading model: “It will not break benchmark records.”
A preview of GPT-4.5 is already available for ChatGPT Pro subscribers. Starting March 2, ChatGPT Plus and Team users will also receive the model. It will also be launched for Enterprise and Edu accounts in the coming weeks.
Sam Altman explained that the company wanted to launch the model for all subscriptions at once, but they ran out of GPUs. "We will add tens of thousands of GPUs next week and roll it out to the plus tier then," Altman wrote in X.