OpenAI has launched GPT-4.5, code-named Orion, marking its largest AI model to date, developed with unprecedented computing power and data. While significant, OpenAI does not classify it as a frontier model.
Starting Thursday, subscribers to ChatGPT Pro can access GPT-4.5 as part of a research preview, with developers on paid API tiers also gaining access. Users of ChatGPT Plus and ChatGPT Team will receive it the following week.
GPT-4.5 was developed using a pre-training phase called unsupervised learning, similar to previous models. It enhances pattern recognition, connection drawing, and creative insight generation. Early tests indicate a more natural interaction experience, improved user intent understanding, and reduced hallucinations.
OpenAI emphasizes two axes of intelligence: unsupervised learning and reasoning. GPT-4.5 exemplifies scaling unsupervised learning, trained on Microsoft Azure AI supercomputers, resulting in broader knowledge and reliability.
Safety Measures
With increased capabilities, safety measures were prioritized. GPT-4.5 was trained using new supervision techniques combined with traditional methods, and extensive safety tests were conducted pre-deployment.
ChatGPT Pro users can select GPT-4.5 across platforms. It supports up-to-date information retrieval, file and image uploads, but lacks multimodal features like Voice Mode. Developers can use GPT-4.5 in various APIs, benefiting from its emotional intelligence and creativity for applications like writing, coaching, and complex task automation.
Cost and Future Considerations
GPT-4.5 is compute-intensive and more expensive than its predecessor, GPT-4o. OpenAI is evaluating its long-term availability in the API while seeking user feedback to guide future developments.
GPT-4.5 represents a significant advancement in unsupervised learning, inviting exploration of its novel capabilities and potential applications.
