OpenAI's GPT-5.4 Can Now Control Your Computer

OpenAI released GPT-5.4 on March 5, and it comes with a capability that changes what AI can actually do for you: it can operate your computer like a human.

What Happened

GPT-5.4 is OpenAI’s newest flagship model, available now in ChatGPT, the API, and their Codex coding tool. It’s the first general-purpose AI model to ship with native computer-use mode built in — not as a bolted-on add-on, but as a core feature.

The model also supports a 1-million-token context window, meaning it can read and reason over roughly 750,000 words in a single conversation. For reference, that’s about six full-length novels at once.

What “Computer Use” Actually Means

When you give GPT-5.4 access to computer-use mode, it takes screenshots of your screen, clicks buttons, types text, opens apps, fills out forms, and moves between windows — all on its own. It uses a “build-run-verify-fix” loop, meaning it checks whether it completed the task correctly before handing back control to you.

The results are striking: on OSWorld, an industry benchmark for desktop navigation, GPT-5.4 scored 75% — compared to 72.4% for human experts. The AI is now statistically better at clicking through software interfaces than people are.

Who Can Use It Right Now

Computer-use mode is currently available to developers and enterprise customers via the API. Everyday ChatGPT users aren’t yet able to point the model at their desktop and say “go”, but OpenAI has indicated broader rollout is coming.

The standard GPT-5.4 model is available to all ChatGPT users, including Plus and Pro subscribers. API pricing starts at $2.50 per million input tokens and $15.00 per million output tokens.

What This Means for You

For most people right now, the practical wins from GPT-5.4 are in the quality of responses — 33% fewer factual errors compared to GPT-5.2, stronger reasoning, and a much larger context window for handling long documents.

The computer-use capability is the bigger shift to watch. Tasks that currently require you to manually open apps, copy-paste between tools, fill out web forms, or click through multi-step processes are exactly what this technology is built for. When it becomes widely accessible, the question won’t be “can AI help me with this?” — it’ll be “should I even open this app myself?”

Think of it like hiring an assistant who never gets tired of clicking through spreadsheets at 2am.