GPT-4.5 is a large language model within OpenAI's GPT series. It was released on February 27, 2025. GPT-4.5 can be accessed by Pro users through the model picker on web, mobile, and desktop, with plans to expand to other tiers. It can also be accessed via the OpenAI API or the OpenAI Developer Playground.^[2]

It was primarily trained using unsupervised learning, though combined with supervised fine-tuning and reinforcement learning from human feedback. It was trained using Microsoft Azure.^[2]

The model is much more expensive than GPT-4o. As of February 2025, it costs $75 per million input tokens and $150 per million output tokens, as opposed to $2.50 per million input tokens and $10 per million output tokens for GPT-4o.^[3]

The model was tested on the MMLU test set, which tested 15 different languages, namely Arabic, Bengali, Chinese, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Spanish, Swahili, and Yoruba. The model outperformed GPT-4o on all 15 languages.^[1]

Cade Metz, writing for New York Times, stated that the model "signifies the end of an era" and was "unlikely to generate as much excitement as GPT-4".^[4] Many other outlets, such as The Verge and Axios, also covered the model's release.^[5][6]

Ethan Mollick, a professor at the Wharton School at the University of Pennsylvania, said he found himself "impressed by GPT-4.5's vision ability".^[7]

Andrej Karpathy tweeted that he found the model "a little bit better" but "not exactly in ways that are trivial to point to".^[8] Simon Willison wrote a blogpost in which he described the model as being "very expensive" and "weirdly slow".^[9] In particular, in one experiment, the model took nearly two minutes to generate a 648-token SVG image.^{[10][unreliable source?]}