OpenAI has unveiled its latest and largest AI language model, GPT-4.5, which is initially available as a research preview for ChatGPT Pro users.

GPT-4.5 demonstrates improvements in writing skills and global knowledge, featuring what OpenAI describes as a "more refined personality" compared to previous models. The company states that interactions with GPT-4.5 will feel more "natural," with enhanced capabilities in pattern recognition and making connections, making it suitable for writing, coding, and "solving real-world problems."

However, OpenAI notes that GPT-4.5 doesn't introduce enough new features to be considered a cutting-edge model. In a leaked document, OpenAI mentioned: "GPT-4.5 isn't a cutting-edge model, but it's OpenAI's largest large language model (LLM), offering over ten times the computational efficiency of GPT-4. Compared to previous reasoning versions, it lacks seven entirely new cutting-edge features and performs below o1, o3-mini, and research-focused models in most readiness evaluations." OpenAI has since removed this reference from the updated document.

Reportedly, OpenAI utilized its o1 reasoning model, codenamed Strawberry, along with synthetic data to train GPT-4.5. The training process employed new supervised techniques combined with traditional methods like Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), similar to the training approach used for GPT-4o.

Despite certain limitations, OpenAI claims that GPT-4.5 generates significantly fewer fabrications than GPT-4o and slightly fewer than the o1 model. OpenAI researcher Raphael Gontijo Lopes stated during a company livestream: "We've tuned GPT-4.5 to be a better collaborative partner, making conversations feel warmer, more intuitive, and emotionally nuanced. To measure this, we had human testers compare it with GPT-4o, and GPT-4.5 excelled in almost all categories."

OpenAI CEO Sam Altman acknowledged on X platform that GPT-4.5 is a "massive and expensive model" that "won't dominate benchmark tests."

OpenAI announced that following its release to Pro users, GPT-4.5 will become accessible to Plus and Team users next week, followed by Enterprise and Edu users. Additionally, the model is now available on Microsoft's Azure AI Foundry platform, alongside new models from Stability, Cohere, and Microsoft.

Previous reports indicated that OpenAI planned to launch GPT-4.5 by the end of February, potentially followed by GPT-5 at the end of May. Sam Altman described GPT-5 as "a system integrating many of our technologies," incorporating OpenAI's new o3 reasoning model, which was previewed in the company's "12 Days of Christmas" announcements last December.

While OpenAI released o3-mini last month, o3 will only be available as part of the upcoming GPT-5 system. This aligns with OpenAI's goal of combining its large language models to ultimately create a more powerful model that could be labeled as Artificial General Intelligence (AGI).