Abdelrahman
OpenAI is set to introduce its latest creation, GPT-4.5, as a research preview today. Positioned as the company’s most extensive and knowledgeable model to date, GPT-4.5 is not classified as a frontier model, and its performance may fall short compared to models like o1 or o3-mini.
With enhanced writing skills and improved overall knowledge, GPT-4.5 aims to offer a more refined user experience compared to its predecessors. However, it does not bring enough groundbreaking features to qualify as a frontier model. OpenAI notes that “GPT-4.5 is not a frontier model, yet it is the largest large language model (LLM) from OpenAI, showing over 10 times the computational efficiency of GPT-4.” According to a recently leaked document, it does not present seven entirely new capabilities compared to earlier reasoning releases and tends to perform below o1, o3-mini, and advanced research on most evaluation metrics.
Reports have suggested that OpenAI leveraged its o1 reasoning model, code-named Strawberry, to use synthetic data in the training of GPT-4.5. The company indicated that this model was trained using a combination of innovative supervision techniques alongside traditional approaches, such as supervised fine-tuning (SFT) and reinforcement learning based on human feedback (RLHF), much like the strategies implemented for GPT-4.
While GPT-4.5 exhibits some limitations, OpenAI claims that it generates fewer inaccuracies than GPT-4o and slightly fewer than its o1 counterpart. Current speculation suggests that access to GPT-4.5 might be restricted to ChatGPT Pro users, similar to the exclusive release of other models.
In a recent article, I revealed that OpenAI aimed to launch GPT-4.5 by the end of February and has plans for GPT-5 as early as late May. OpenAI’s CEO, Sam Altman, referred to GPT-5 as a “system that integrates a considerable amount of our technology.” This upcoming model promises to include OpenAI’s innovative o3 reasoning framework, which was hinted at during the company’s 12 Days of Christmas announcements.
Although OpenAI launched o3-mini last month, it has chosen to include o3 exclusively in the upcoming GPT-5 system. This strategy aligns with the company’s vision to consolidate its large language models in pursuit of developing a robust model that could ultimately be recognized as artificial general intelligence (AGI).
