OpenAI introduces GPT-4.5, its newest and largest non-reasoning AI model

OpenAI has unveiled GPT-4.5, its largest and most knowledgeable non-reasoning language model to date, surpassing the capabilities of its predecessor, GPT-4o. While not classified as a frontier model, GPT-4.5 boasts enhanced world knowledge, improved writing, and a refined interaction style. It excels in writing, programming, and problem-solving, with improved pattern recognition and conversational abilities that early testers describe as more emotionally intelligent and engaging.

GPT-4.5 supports features like search, file and image uploads, and canvas, but lacks native multimodal capabilities such as Voice Mode, video, and screen sharing at launch. It is also compute-intensive, which may affect deployment and scalability. In benchmark tests, GPT-4.5 scored a 38% on the SWE-bench Verified benchmark, showing a 2-7% improvement over GPT-4o. It also demonstrates significant reductions in hallucinations compared to previous models, enhancing its reliability. However, GPT-4.5 is not intended to replace models o3-mini or deep research, serving instead as a more general-purpose model.

The research preview of GPT-4.5 is now available for ChatGPT Pro users through the model picker on web, mobile, and desktop, with availability expanding to ChatGPT Plus and Teams users next week.