Home / Technology / GPT-5.4: AI Now Does, Not Just Tells
GPT-5.4: AI Now Does, Not Just Tells
6 Mar
Summary
- New GPT-5.4 can execute commands like clicking and editing files.
- Model operates on PCs via API/Codex, not within ChatGPT interface.
- Represents a leap to AI agents controlling PCs with user direction.
OpenAI has launched GPT-5.4, a new flagship model that moves beyond traditional conversational AI. This advanced model can now execute commands on a user's computer, a significant departure from AI that merely provides instructions.
GPT-5.4's ability to interact with a PC includes issuing commands for mouse clicks, editing files, and interpreting screenshots. This functionality is primarily available when accessed through the OpenAI API or the Codex coding tool. These platforms allow GPT-5.4 to act as an AI agent, controlling computer programs and browsing the web.
When used via ChatGPT, however, GPT-5.4 remains within its chat interface and integrated applications. Despite this limitation, the API and Codex integration represents a substantial leap, building upon earlier Codex-specific models. It enables complex autonomous tasks, such as managing financial software, under user supervision.
