OpenAI’s new GPT-5.4 model is a big step toward autonomous agents
The new model can operate your computer, issuing keyboard commands and writing code to complete tasks across applications.
OpenAI has launched GPT-5.4, marking a significant step toward its vision of autonomous AI agents. The model is the company's first with native computer use capabilities, meaning it can directly operate a computer—issuing keyboard and mouse commands and writing code—to complete tasks across various applications. This positions GPT-5.4 as a foundational tool for the 'agentic future' where AI works in the background on complex, multi-step jobs. Alongside this, OpenAI is rolling out GPT-5.4 Thinking to ChatGPT, which provides reasoning outlines for complex queries, allowing users to tweak requests mid-process.
The technical advancements are substantial: OpenAI claims GPT-5.4 is its 'most factual model yet,' with individual claims 33% less likely to be false compared to GPT-5.2. It shows improved performance in web browsing, tool/API calling, and can more persistently search across multiple sources to synthesize answers. The model is available now across ChatGPT, Codex, and the API. GPT-5.4 Pro, for maximum performance on complex tasks, is available in the API and for ChatGPT Enterprise and Edu users, signaling a clear push for professional and developer adoption.
- First OpenAI model with native computer control, can issue keyboard/mouse commands and write code to operate applications.
- Claims to be 33% more factual than GPT-5.2, with improved multi-source search and synthesis for complex questions.
- GPT-5.4 Thinking model provides reasoning outlines in ChatGPT, allowing users to guide responses without starting over.
Why It Matters
Moves AI from a conversational tool to an active workforce that can automate complex digital tasks across your device.