OpenAI has officially launched GPT-4.5, its latest and largest AI language model, initially available as a research preview for ChatGPT Pro users.
Described as the company’s “most knowledgeable model yet,” GPT-4.5 boasts enhanced capabilities but is not classified as a frontier model, meaning it may not outperform models like o1 or o3-mini.
According to OpenAI, GPT-4.5 delivers improved writing, deeper world knowledge, and a more refined conversational experience. The model is designed to feel more natural, with better pattern recognition and problem-solving skills, making it highly suitable for tasks like writing, coding, and practical reasoning.
Despite these improvements, OpenAI acknowledges that GPT-4.5 does not introduce enough groundbreaking advancements to be considered a frontier model. A leaked document initially stated that while GPT-4.5 is over ten times more computationally efficient than GPT-4, its performance remains below that of o1 and o3-mini in some areas. This reference was later removed from OpenAI’s official announcement.
Reports indicate that OpenAI used its o1 reasoning model, codenamed “Strawberry,” to train GPT-4.5 with synthetic data. The model was developed using a combination of new supervision techniques and traditional methods like supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), similar to the approach used for GPT-4o.
A key advantage of GPT-4.5 is its reduced tendency to generate hallucinated responses, outperforming GPT-4o in this regard and slightly improving upon the o1 model. OpenAI researcher Raphael Gontijo Lopes noted that the model has been optimized for a more intuitive and emotionally nuanced conversation experience. “We aligned GPT-4.5 to be a better collaborator, making interactions feel warmer and more natural,” Lopes stated during OpenAI’s livestream. Human testers reportedly found GPT-4.5 superior to GPT-4o across multiple categories.
OpenAI CEO Sam Altman described GPT-4.5 as a “giant, expensive model” in a post on X, acknowledging that it “won’t crush benchmarks.”
After its initial release for Pro users, OpenAI plans to roll out GPT-4.5 to Plus and Team users next week, followed by Enterprise and Edu users. The model is also available on Microsoft’s Azure AI Foundry platform, alongside new models from Stability, Cohere, and Microsoft.
Recent reports suggested OpenAI was targeting a late February release for GPT-4.5 and is planning to launch GPT-5 as early as late May. Altman has described GPT-5 as a “system that integrates a lot of our technology,” featuring OpenAI’s upcoming o3 reasoning model, which was previewed during its “12 Days of Christmas” announcements in December.
While OpenAI recently introduced o3-mini, the full o3 model will debut as part of GPT-5. This aligns with OpenAI’s broader goal of merging its AI models to develop a more advanced system approaching artificial general intelligence (AGI).
