The newest version features built-in capabilities for computer use, allowing it to perform tasks across your device and applications. OpenAI is releasing GPT-5.4, the latest AI version combining improvements in reasoning, coding, and professional tasks involving spreadsheets, documents, and presentations. It’s the first OpenAI model with built-in computer use capabilities, enabling it to control a computer and execute tasks across various applications.
The model represents a move toward an agentic future, where AI-powered networks perform complex tasks online and in software. OpenAI introduced ChatGPT Agent alongside various agentic tools last year, which can take control of your computer to execute tasks like sourcing and purchasing meal ingredients.
OpenAI is integrating GPT-5.4 into its API and Codex, its AI-driven coding tool, and deploying its reasoning version, GPT-5.4 Thinking, to ChatGPT. GPT-5.4 can write code to operate computers, issue keyboard and mouse commands from screenshots, and demonstrate improved web browsing capabilities, accurately employing tools and APIs to finalize tasks.
The model excels in answering complex questions by gathering information from multiple sources, persistently searching to find the most relevant data, particularly for difficult questions, and synthesizing it into a coherent response. OpenAI asserts GPT-5.4 is its “most factual model yet,” reducing false claims by 33 percent compared to GPT-5.2.
Within ChatGPT, GPT-5.4 Thinking will deliver a work outline for complex inquiries, enabling users to adjust their requests without starting over. This feature, now available in the ChatGPT web app and Android, is “coming soon” to iOS.
GPT-5.4 is deploying across ChatGPT, Codex, and the API, with the GPT-5.4 Thinking model available to Plus, Team, and Pro users. There’s also a GPT-5.4 Pro model for “maximum performance on complex tasks” available in the API, ChatGPT Enterprise, and Edu users.