In mid-February, OpenAI CEO Sam Altman had already revealed GPT-4.5 will be released in a few weeks.Earlier, in OpenAI’s livestream, GPT-4.5 was officially unveiled, showing significant improvements over GPT-4o. Sam Altman even called this GPT-4.5 a massive and expensive model. It is currently available to all ChatGPT Pro members, while paid members such as Plus and Enterprise are expected to get access next week.
OpenAI GPT-4.5 model officially launches, with significantly fewer hallucinations than GPT-4o.
GPT-4.5 is the successor to GPT-4o and belongs to the GPT series. According to OpenAI, the research preview of GPT-4.5 is the most powerful and outstanding model in the conversational domain to date.
Based on early test results, GPT-4.5 feels more fluid and natural in conversation, has a broader knowledge base, understands user intent more accurately, and effectively assists with writing, programming, and solving practical problems.
More notably, GPT-4.5 is fundamentally built on expanded unsupervised learning techniques, leveraging more powerful computing resources, larger datasets, and improved architecture and optimization to achieve breakthrough progress, making AI more accurate, intelligent, and closer to human cognition.
OpenAI also believes that GPT-4.5 will have a lower probability of generating false information. The figure below shows test data from simple Q&A. In terms of hallucination rate, GPT-4.5 is only 37.1%, much lower than GPT-4o’s 61.8%:
As for accuracy, GPT-4.5 performed the best at 62.5%, while GPT-4o only had 38.2%. This shows that when asking relatively simple questions, GPT-4.5 can give the best answers.
GPT-4.5 also shows significant improvements in collaboration, comprehension, and creativity, allowing it to more accurately grasp human needs and make conversations more natural and intuitive, while also providing better assistance in creative fields such as writing and design.
You can refer to the GPT-4.5 vs GPT-4o 2 test data in the figure below. In daily queries, GPT-4.5 has a win rate of 57.0%, professional queries reach 63.2%, and creative intelligence also has 56.8%;
Unlike reasoning models such as o1, GPT-4.5 does not engage in thinking when answering questions. OpenAI states that compared to OpenAI o1 and OpenAI o3-mini, GPT-4.5 is a more general-purpose and inherently more intelligent model.
Sam Altman said on the X social platform: “GPT-4.5 is a huge and expensive model. We originally hoped to roll it out to both Plus and Pro users at the same time, but because user growth has been too fast, our GPU supply is insufficient. However, next week we will add tens of thousands of GPUs, and we will push it to Plus tier users then.”
GPT-4.5 is ready!
good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i've sat back in my chair and been astonished at getting actually good advice from an AI.
bad news: it is a giant, expensive model. we…
— Sam Altman (@sama) February 27, 2025
Although GPT-4.5 appears to be a significant improvement over GPT-4, strictly speaking, it’s only a transitional model, because Sam Altman has already said that GPT-5 will be released in a few months, and it will incorporate reasoning models into a single unified model to make it easier for users to use.
How to use GPT-4.5 in ChatGPT
Starting today, ChatGPT Pro users can select the GPT-4.5 model in the model selector on web, mobile, and desktop.
- Starting next week, we’ll gradually roll this out to Plus and Team users.
- The week after that, it will roll out to Enterprise and Edu users.
In addition, GPT-4.5 supports search, file and image uploads, and Canvas, but multimodal features like voice mode, video, and screen sharing are not yet available.
Source:OpenAI
Source: KOCPC Chinese