OpenAI has unveiled GPT-4.5, its most advanced and powerful model to date. The company has launched a research preview of GPT-4.5, marking a significant step in scaling up pre- and post-training. The latest model enhances its ability to recognise patterns, establish connections, and generate creative insights, though without deep reasoning. Early testing indicates that interactions with GPT-4.5 feel more natural. It boasts a broader knowledge base, improved alignment with user intent, and greater emotional intelligence (EQ), making it particularly effective for tasks such as refining writing, coding, and solving practical problems.
GPT-4.5 as a Research Preview
“We also expect it to hallucinate less. We’re sharing GPT-4.5 as a research preview to understand its strengths and limitations better,” OpenAI stated.
“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.”
The company has advanced AI capabilities by scaling two key paradigms: unsupervised learning and reasoning.
Scaling Reasoning and Unsupervised Learning
OpenAI has found that teaching AI to think step by step before responding helps it handle complex STEM and logic problems more effectively. Models like OpenAI o1 and OpenAI o3-mini follow this approach, making them better at reasoning. At the same time, unsupervised learning helps AI build a deeper understanding of the world and sharpens its intuition.
GPT-4.5 represents an advancement in unsupervised learning by leveraging increased computational power, data, and architectural innovations.
Broader Knowledge and Deeper Understanding
As a result, GPT-4.5 possesses a broader knowledge base and a deeper understanding of the world, reducing hallucinations and improving reliability across a range of topics.
“As we scale our models to solve more complex problems, it’s increasingly important to enhance their understanding of human needs and intent,” OpenAI noted.
To achieve this, the company developed new, scalable techniques that enable the training of larger, more powerful models using data derived from smaller ones. These innovations enhance GPT-4.5’s ability to follow instructions, grasp nuances, and engage in more natural conversations.
OpenAI also highlighted that GPT-4.5 has an improved ability to interpret subtle cues and implicit expectations, leading to greater nuance and emotional intelligence. Additionally, the model demonstrates a stronger sense of aesthetics and creativity, excelling in writing and design-related tasks.
Sam Altman, OpenAI’s CEO, expressed his excitement about the model, stating:
“GPT-4.5 is ready! The best part? It’s the first model that feels like talking to a thoughtful person. I’ve had several moments where I sat back in my chair, genuinely astonished by the quality of advice I received from an AI.”
However, Altman also acknowledged that GPT-4.5 is a large and expensive model.
“We wanted to launch it to both Plus and Pro users at the same time, but we’ve been growing rapidly and have run out of GPUs. We’ll be adding tens of thousands of GPUs next week and rolling it out to the Plus tier then—hundreds of thousands more are coming soon, and I’m pretty sure you’ll use everyone we deploy,” he added.