OpenAI Unveils GPT-5.4, Its First Model With Native Computer Use
Model Overview
OpenAI announced GPT-5.4 as the latest iteration of its generative‑pretrained transformer series. The model integrates improvements across three core areas: reasoning, code generation, and handling professional‑office workflows such as spreadsheets, documents, and presentations. OpenAI describes GPT-5.4 as its most factual model to date, noting that individual claims are 33 percent less likely to be false compared with the previous GPT-5.2 version.
Native Computer Use Capabilities
For the first time, an OpenAI model can operate a computer directly. GPT-5.4 can interpret screenshots, generate code that controls applications, and issue keyboard and mouse commands to complete tasks across multiple software environments. This capability extends to web browsing, where the model can navigate pages, call tools and APIs more accurately, and synthesize information from disparate sources.
Agentic Future and ChatGPT Integration
OpenAI frames GPT-5.4 as a step toward an "agentic" future, where a network of AI‑powered agents works in the background to accomplish complex online jobs. The company introduced a ChatGPT Agent that, like other emerging tools, can take control of a user's computer to perform actions such as searching for and purchasing meal ingredients.
Within ChatGPT, the specialized GPT-5.4 Thinking model provides an outline of its reasoning process for complex queries. Users can adjust or refine their requests mid‑response, avoiding the need to restart a conversation. This interactive feature is already live in the ChatGPT web app and on Android, with iOS support slated for the near future.
Availability and Target Users
GPT-5.4 is rolling out across OpenAI’s platforms. The base model is accessible via the API and the Codex coding tool. The GPT-5.4 Thinking variant is available to Plus, Team, and Pro subscribers of ChatGPT. Additionally, a GPT-5.4 Pro version, designed for maximum performance on demanding tasks, is being released for API customers as well as ChatGPT Enterprise and Education users.
Impact on Users and Developers
The new model’s ability to directly manipulate computers opens a range of possibilities for developers and enterprises seeking automation solutions. By combining stronger factual grounding, multi‑source information synthesis, and hands‑on computer control, GPT-5.4 aims to streamline workflows that previously required manual intervention or separate scripting tools.
Used: News Factory APP - news discovery and automation - ChatGPT for Business