OpenAI has introduced the latest iteration of its AI model, GPT-5.4, marking a significant advancement in the realm of artificial intelligence. This new model is designed to excel in reasoning, coding, and professional tasks like working with spreadsheets, documents, and presentations. Notably, GPT-5.4 stands out as OpenAI’s first model with native computer usage capabilities, empowering it to undertake tasks autonomously across various applications.
One of the key features of GPT-5.4 is its ability to interact with computers directly, enabling it to execute commands, operate applications, and respond to screenshots. OpenAI is integrating GPT-5.4 into its API and Codex, the AI-driven coding tool, while also deploying its reasoning model, GPT-5.4 Thinking, to enhance ChatGPT. This new model showcases improved performance in web browsing tasks and demonstrates enhanced accuracy in utilizing tools and APIs to efficiently accomplish designated tasks.
OpenAI emphasizes that GPT-5.4 excels in processing multifaceted queries that demand information aggregation from diverse sources. The model’s capability to search persistently across multiple rounds to extract pertinent data, especially for complex inquiries, sets it apart. Furthermore, GPT-5.4 is described as OpenAI’s most factually reliable model to date, with a significantly reduced likelihood of false claims compared to its predecessor, GPT-5.2.
Within ChatGPT, the introduction of GPT-5.4 Thinking enhances user interactions by providing detailed work outlines for intricate requests. Users can now refine their queries during the model’s response, facilitating a more precise outcome without restarting the process. This feature is presently accessible on the ChatGPT web app and Android platforms, with an imminent rollout planned for the iOS app.
Source: The Verge