OpenAI has announced the launch of GPT-4.5, a new language model that has already generated significant interest. The model, codenamed Orion, is the largest in the company’s history, built using more computational power and data than any of its predecessors. Since its launch, ChatGPT Pro subscribers can already try out the new model’s capabilities as part of a research preview.
GPT-4.5 features improved writing abilities, a significant expansion of world knowledge, and higher emotional intelligence. It also demonstrates more natural interaction with users, making it suitable for solving practical tasks, writing, and programming. Nevertheless, OpenAI emphasizes that GPT-4.5 is not a “frontier” model, as it does not outperform some of the company’s existing models, such as o1 and o3-mini.
The model has received new training methods, including scalable techniques that allow for better control and understanding of human intent. Although GPT-4.5 does not yet support multimodal features such as voice or video input, it offers file and image upload capabilities, as well as enhanced planning features for complex workflows.
OpenAI highlights the safety of the new model, combining traditional reinforcement learning with human feedback. Despite higher computational requirements and costs compared to previous models, the company will assess the long-term availability of GPT-4.5 based on user feedback.
