OpenAI officially released its large language model named GPT-4.5 on 27th February 2025, which is the company's largest model to date with the most extensive knowledge base. GPT-4.5 operates with significantly fewer hallucinations (37.1% compared to GPT-4o's 61.8%) and is primarily available to ChatGPT Pro subscribers and developers through API access.
The GPT-4.5 model has improved computational efficiency tenfold compared to previous models, and has combined new supervisory techniques with traditional methods such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF). Sam Altman, CEO of OpenAI, commented on social media: "the first model that, to me, feels like talking to a thinking person. I've had several moments where I sat back in my chair, astonished that I'm actually receiving good advice from an AI model." According to the company, the model possesses more advanced emotional intelligence, better follows user intent, and enables more naturally flowing conversations.
GPT-4.5 demonstrates significant advancement compared to GPT-4o, particularly in thinking and reasoning capabilities, as evidenced by a 20 percentage point improvement in the GPQA (Graduate-Level Google-Proof Q&A Benchmark) test. Based on OpenAI's internal assessments, it has surpassed its predecessor in all evaluated categories, including writing skills, programming, and practical problem-solving. The new model provides more accurate responses, better understands complex instructions, and more efficiently solves logical tasks. These improvements create opportunities for next-generation artificial intelligence applications that offer unprecedented levels of accuracy and reliability. OpenAI aims to further expand AI applications in education, research, and industrial innovation through GPT-4.5.
Sources:
1.
2.
Carl Franzen
3.
Tom Warren
4.
Peter Gostev
