OpenAI has released GPT-4.5, the company’s freshest and most extensive AI language model, though it will initially be available as a research preview.
OpenAI is calling GPT-4.5 its “most knowledgeable model yet.” However, the company has cautioned that it is not a “frontier” model and might not outperform the o1 or o3-mini models. This release is currently available for ChatGPT Pro users.
GPT-4.5 boasts improved writing capabilities and better world knowledge and has been given what OpenAI calls a “refined personality.” The company claims that interacting with GPT-4.5 will feel more natural as the model can better recognize patterns and make connections, making it suitable for writing, programming, and problem-solving.
However, in documentation that was leaked before the official announcement, OpenAI highlighted that the new technology doesn’t introduce enough new capabilities to be considered a frontier model. The mention of the model’s limitations was later removed from an updated version of the document.
GPT-4.5 is not a frontier model, but it is OpenAI’s largest LLM, improving on GPT-4’s computational efficiency by more than 10x.
It was reported that OpenAI used its o1 reasoning model, codenamed Strawberry, to train GPT-4.5 using synthetic data. The company stated that it trained GPT-4.5 using new supervision techniques combined with traditional methods, such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), similar to those employed for GPT-4o.
Despite its limitations, OpenAI claims that GPT-4.5 hallucinates considerably less than GPT-4o, and also slightly less than its o1 model. Raphael Gontijo Lopes, a researcher at OpenAI, noted during the company’s livestream that GPT-4.5 was designed to be a better collaborator, enhancing conversations by making them feel warmer, more intuitive, and more emotionally nuanced. He added that human testers evaluated the model against GPT-4o, with GPT-4.5 outperforming it in nearly every category.
In a post on X, OpenAI CEO Sam Altman acknowledged that the new version is a “giant, expensive model” and that it “won’t crush benchmarks.”
Following the launch 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 currently accessible on Microsoft’s Azure AI Foundry platform alongside new models from Stability, Cohere, and Microsoft.
OpenAI also revealed plans to launch GPT-5, which could happen by late May. Sam Altman characterized GPT-5 as a “system that integrates a lot of our technology” and will include OpenAI’s new o3 reasoning model, which OpenAI previewed during its December announcements.
While OpenAI released o3-mini last month, the o3 model will only be shipped as part of the upcoming GPT-5 system, which is consistent with OpenAI’s aim to combine its large language models to create an artificial general intelligence, or AGI.
