OpenAI unveiled GPT‑5.4—a leading LLM with enhanced computer work skills in agent mode

OpenAI launches GPT‑5.4 – a model capable of controlling a computer

OpenAI announced the release of a new version of its artificial intelligence—GPT‑5.4. According to developers, this model combines enhanced logical reasoning and programming skills with the ability to work with text documents, spreadsheets, and presentations. What is especially important is that GPT‑5.4 now supports direct interaction with the user’s computer and various applications.

What’s new in GPT‑5.4
Feature | Description
---|---
PC Control | The model can generate code to automate tasks on a computer, emulate mouse and keyboard actions on request (in the form of screenshots).
Browser and API | Improved interaction with browsers and third‑party services via their APIs.
Multi‑search | For complex questions GPT‑5.4 conducts several search sessions to gather data from different sources, then synthesizes it into an understandable answer.
Error reduction | OpenAI claims that the number of actually incorrect statements has dropped by 33 % compared with GPT‑5.2, making the model “the most reliable so far.”

Reasoning version – GPT‑5.4 Thinking
The new “reasoning” submodel, available in ChatGPT as GPT‑5.4 Thinking, presents a solution plan immediately after receiving a request. The user can adjust the prompt on the fly, eliminating the need to start over and simplifying the attainment of the desired result.

Where it’s already usable
Platform | Availability | API
---|---|---
Fully available (including the powerful GPT‑5.4 Pro for Enterprise/Edu).
Codex | Support for the new model in the code app.
ChatGPT | Base GPT‑5.4 is already deployed; Thinking is available to Plus, Team, and Pro subscribers.

Thus, OpenAI takes a step toward widespread deployment of AI agents that can act on behalf of users, perform complex online tasks, and control software on their computers.