OpenAI’s new flagship GPT mannequin can management your PC
Abstract created by Sensible Solutions AI
In abstract:
- OpenAI’s new GPT-5.4 flagship mannequin introduces agentic AI capabilities that may immediately management PC features like clicking, file modifying, and spreadsheet administration.
- PCWorld reviews this improvement marks a major shift towards autonomous AI brokers performing pc duties with out human intervention.
- The superior options are accessible by means of OpenAI API and Codex platforms, representing a significant leap past conventional conversational AI interactions.
Bear in mind when AI fashions might solely inform you what to do? Now, the most recent LLMs can truly do issues with the assistance of agentic AI software program, and OpenAI’s new flagship mannequin is the latest of the bunch.
GPT-5.4 is out now on ChatGPT (the place it goes by the identify GPT-5.4 Considering) in addition to on the OpenAI API and OpenAI’s coding instrument Codex (a model of which simply got here out for Home windows).
This new GPT arrives with a variety of new and revamped methods, beginning with its improved spreadsheet abilities, extra environment friendly reasoning (which means it may resolve issues utilizing fewer tokens, thus costing you much less), and skill to point out you an “upfront” plan earlier than executing advanced duties, supplying you with an opportunity to steer the mannequin in a brand new path earlier than it will get to work.
Most curiously, GPT-5.4 marks OpenAI’s first general-purpose mannequin that may truly do issues in your pc, not simply inform you how to do issues. For instance, GPT-5.4 can click on a mouse—or to be extra exact, it may challenge a “click on the mouse” command to an AI agent system in your PC, which does the precise clicking. GPT-5.4 may edit recordsdata in your system, sort keyboard instructions, and “see” screenshots (permitting it to make use of an online browser or work together with pc packages).
Now, an essential caveat right here: GPT-5.4 can solely take cost of your PC when it’s working through the OpenAI API or OpenAI’s Codex instrument. If you’re utilizing GPT-5.4 Considering by means of ChatGPT—that’s, the ChatGPT desktop app or net interface—the LLM continues to be confined to its chatbox and its varied ChatGPT integrations, reminiscent of for Google Drive, Spotify, Adobe Photoshop, and others.
It’s additionally value noting that whereas GPT-5.4 is the primary general-purpose GPT that may truly use your PC, it’s not the primary GPT ever that may achieve this. There have been Codex-specific GPTs that may execute instructions, edit recordsdata, and (to an extent) navigate graphical interfaces and weave their approach by means of net workflows. However with its potential to truly browse the online and take cost of PC packages, GPT-5.4 takes the “computer-use” capabilities of earlier Codex-specific fashions to the following stage.
Which means you can conceivably ask a GPT-5.4-controlled AI agent in your pc to “stability my books on Quicken” and it could be capable to autonomously launch the Quicken app, click on its approach across the interface, and stability your accounts.
After all, whether or not you’d need GPT-5.4 messing round in Quicken by itself is a separate query altogether. For delicate duties, you’d possible wish to be trying over its shoulder as it really works, as you are able to do whereas coding with GPT-5.4 within the Codex app.
Nonetheless, the “do, don’t simply inform” capabilities of GPT-5.4 function an ideal instance of the place we’re headed: AI agent-controlled PCs which can be doing issues on their very own, with high-level path from us. That stated, getting our AI brokers to comply with our instructions accurately would be the actual trick.

