OpenAI debuts GPT-4.5, its biggest model yet
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Natalie Peeples/Axios
OpenAI on Thursday released an early version of GPT-4.5, a significant update to the large language model underlying ChatGPT that OpenAI says will be better at recognizing patterns and drawing connections.
Why it matters: The release marks the next evolution of OpenAI's flagship model lineage that has been an industry front-runner since the start of the generative-AI boom.
Driving the news: This is OpenAI's largest model yet — though the company declined to offer details about its size or the computing resources it took to train it.
- GPT-4.5 should hallucinate less, follow instructions better and deliver interactions that feel more natural.
- The "research preview" version of GPT 4.5 that was released Thursday supports real-time search and image and file uploads but not voice mode, video or screen sharing.
- GPT-4.5 will be made available starting today for subscribers of the $200-per-month ChatGPT Pro and next week for other paid subscribers, including the $20-per-month ChatGPT Plus service. It's also being made available to those who pay to use OpenAI's API.
Yes, but: OpenAI noted that because it is a larger model, GPT-4.5 is more expensive to deliver.
- "GPT‑4.5 is a very large and compute-intensive model, making it more expensive than and not a replacement for GPT‑4o," it said.
What they're saying: OpenAI CEO Sam Altman wrote in a post on X, "Good news: it is the first model that feels like talking to a thoughtful person to me... this isn't a reasoning model and won't crush benchmarks. It's a different kind of intelligence and there's a magic to it I haven't felt before."
Between the lines: Until last year, much of the progress in large language models came from increasing the size of the underlying model and the amount of data used to train it.
- But that approach, known as "unsupervised learning" or "pretraining," has begun to offer diminishing returns. So the focus has switched to "reasoning" — that is, letting models take more time and use more computing resources at the time of a query to deliver a better answer.
GPT-4.5 represents OpenAI's first major move in some time at improving its AI using the former approach, scaling up the pretraining phase.
- OpenAI says the result is that GPT-4.5 has more and broader world knowledge and should have increased reliability.
What they're saying: "We're sharing GPT‑4.5 as a research preview to better understand its strengths and limitations," OpenAI said in a blog post. "We're still exploring what it's capable of and are eager to see how people use it in ways we might not have expected."
What's next: OpenAI hinted it will add reasoning improvements on top of GPT-4.5.
- The company's "roadmap" announcement earlier this month, pointing the way toward the eventual release of GPT-5, called 4.5 "our last non-chain-of-thought model."
- "We believe reasoning will be a core capability of future models, and that the two approaches to scaling — pre-training and reasoning — will complement each other," OpenAI said. "As models like GPT‑4.5 become smarter and more knowledgeable through pre-training, they will serve as an even stronger foundation for reasoning and tool-using agents."
