(Reuters) – OpenAI, the developer behind ChatGPT, announced on Thursday the introduction of a research preview of their latest GPT-4.5 model for Pro users and developers globally, with a plan to broaden availability in the near future. GPT-4.5 boasts enhanced capabilities in recognizing patterns, generating innovative insights sans logical reasoning, and displaying heightened emotional intelligence, as per the company’s statement. Backed by Microsoft, OpenAI disclosed intentions to initiate the rollout of GPT-4.5 to ChatGPT Plus and Team users in the upcoming week, followed by Enterprise and Edu users in the subsequent week. CEO Sam Altman described the model as “a colossal, costly model,” attributing the staggered rollout to a shortage in GPU capacity, necessitating the addition of tens of thousands of GPUs in the forthcoming week. Altman further explained that while this operational approach was not ideal, predicting growth surges leading to GPU scarcities posed challenges. OpenAI highlighted that GPT-4.5 exhibited a lower rate of “hallucinations,” addressing a prevalent issue wherein sizeable language models generate false information. The hallucination rate of GPT-4.5, at 37.1%, contrasts with the rates of 61.8% and 44% observed in GPT-4o and its o1 reasoning model, respectively. Noteworthy features of GPT-4.5 include support for file and image uploads, as well as functionality in writing and coding projects, albeit lacking current support for advanced features like voice and video integration. (Reporting by Deborah Sophia in Bengaluru; Editing by Shounak Dasgupta)
