OpenAI today announced a research preview of GPT-4.5, which it describes as its largest and best AI chat model yet. It’s available now to ChatGPT Pro subscribers and developers.
“GPT‑4.5 is a step forward in scaling up pretraining and post-training,” OpenAI says. “By scaling unsupervised learning, GPT‑4.5 improves its ability to recognize patterns, draw connections, and generate creative insights without reasoning.”
In early tests, GPT-4.5 “feels more natural,” the company claims, thanks to its “broader knowledge base, improved ability to follow user intent, and greater ‘EQ,'” a term that is never explained. It’s useful for improving writing, programming, solving practical problems, and other similar tasks, and OpenAI says it should be more reliable and hallucinate less than previous models.
OpenAI is using the release preview period to better determine the model’s strengths and weaknesses and see how customers use it in ways it didn’t anticipate. It will roll out GPT-4.5 to a broader audience later.
But this model is in many ways the end of an era: CEO Sam Altman said recently that this will be the last model OpenAI releases that doesn’t include reasoning capabilities. Future models will combine the traditional GPT models with its newer generation o-series models so that AI will “just work” and not require users to choose a model each time they make a query. The first of those will be called GPT-5 and is expected later this year.