February 27, 2025 — OpenAI today officially launched its latest language model, GPT-4.5, initially available as a research preview to ChatGPT Pro subscribers ($200 monthly subscription). Hailed as the "largest and most knowledgeable" model to date, it builds upon GPT-4o, showcasing significant improvements in reasoning, emotional intelligence, and practical functionalities like writing and coding. OpenAI stated that GPT-4.5 will gradually roll out to Plus, Team, Enterprise, and Edu users in the coming weeks, marking another milestone in artificial intelligence technology.
Key Features: Smarter and More Natural
GPT-4.5 demonstrates remarkable progress across various domains. According to OpenAI's system card, the model, trained through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), significantly reduces "hallucinations" (incorrect or fabricated information) and enhances emotional intelligence and the naturalness of user interaction. For instance, in the PersonQA test, GPT-4.5's hallucination rate is only 19%, significantly lower than GPT-4o's 52%. In terms of user experience, GPT-4.5 excels in writing and programming tasks, generating more creative and aesthetically intuitive content while accurately understanding user intent. X user Simon Willison, after testing it, commented, "It's better at knowledge depth and content generation (like drawing pelicans), and the interaction feels more natural." Furthermore, the model supports file and image uploads and integrates ChatGPT's canvas tool, but currently doesn't support multi-modal features like voice, video, or screen sharing.
Data and Security: Transparency and Responsibility
OpenAI revealed that GPT-4.5's training data comes from diverse sources, including public data, proprietary partner data, and internally curated datasets, rigorously filtered to ensure quality and protect privacy. On the security front, the model is equipped with a Moderation API and classifier to effectively prevent the generation of harmful content. In third-party evaluations, Apollo Research and METR tested GPT-4.5's deception risk and autonomy, revealing an overall "moderate" risk. Risks were deemed moderate in chemistry, biology, radioactive nuclear materials (CBRN), and persuasiveness, while risks in cybersecurity and model autonomy were lower. OpenAI has implemented measures such as pre-training filtration of CBRN data and monitoring for potential misuse to ensure safety.
User Feedback: Enhanced Capabilities, but Speed Needs Optimization
Initial user feedback shows widespread praise for GPT-4.5's breadth of knowledge and creativity. X user @pckraftwerk commended its "significantly improved writing and coding capabilities, with reduced hallucinations." However, speed issues have also raised concerns. Simon Willison noted that when used via API, GPT-4.5's response speed is slower, potentially impacting real-time applications. He also revealed that the model used 10 times more training compute resources than GPT-4, highlighting the significant investment behind its performance improvements.
Furthermore, OpenAI explicitly stated that GPT-4.5 is not designed for deep reasoning, suggesting users opt for o1 or o3-mini models for complex reasoning tasks—a clarification that has sparked some discussion.
Market Impact and Future Outlook
As OpenAI's latest creation, GPT-4.5 is expected to drive productivity innovations in writing, programming, and problem-solving. Its multilingual performance is also impressive, achieving 0.896 accuracy in English, 0.8840 in Spanish, and 0.8598 in Arabic on the MMLU benchmark, surpassing GPT-4o across the board. Industry experts predict that as GPT-4.5 is gradually rolled out, its powerful features and enhanced security will further solidify OpenAI's leading position in the AI field. In the future, OpenAI plans to optimize the model based on user feedback and potentially expand functionalities to include voice and video, further enriching its application scenarios.
Conclusion
The release of GPT-4.5 not only showcases OpenAI's ambition in technology but also highlights its commitment to safety and user experience. Although speed issues remain to be addressed, this model undoubtedly opens a new chapter in AI development. As OpenAI stated in its announcement: "We hope GPT-4.5 will provide users with a smarter and more reliable assistant experience."
