GPT-5.4 Is Here, and It Wants to Do Your Spreadsheets

OpenAI just dropped a new model, and this one isn’t chasing casual users. GPT-5.4 is built squarely for professionals who live inside data, code, and complex workflows.

It’s the company’s most capable frontier model yet for coding and data analysis. And if you’ve ever wished your AI assistant could actually navigate your desktop instead of just talking about it, GPT-5.4 has something new to offer.

Native Computer-Use Puts GPT-5.4 in the Driver’s Seat

This is the first OpenAI model built with native computer-use capabilities from the ground up. That’s a bigger deal than it sounds.

Previous models could attempt to control a desktop, but they were clunky about it. GPT-5.4 issues mouse and keyboard commands far more smoothly than its predecessor. So instead of just generating instructions, it can actually carry out tasks across multiple apps at the same time.

Think of it like the difference between a GPS giving you directions and a self-driving car doing the driving. One tells you what to do. The other just does it.

Thinking Mode Gets a Smarter Default

When you use ChatGPT’s Thinking mode, GPT-5.4 is now the default model powering it. And OpenAI made that mode more collaborative in a meaningful way.

Before diving into a response, the system lays out its plan. You can see how it intends to tackle your request and redirect it before it gets too far down the wrong path. That’s genuinely useful for complex, multi-step tasks where early assumptions can spiral into wasted time.

Plus, web research got a noticeable upgrade. GPT-5.4 handles highly specific queries better than GPT-5.2 did, which matters when you’re hunting for precise technical information rather than broad overviews.

GPT-5.4 Makes Fewer Mistakes Than Its Predecessor

OpenAI claims GPT-5.4 is its most factual model to date. Compared to GPT-5.2, it’s 18 percent less likely to generate a response containing errors.

That might sound like a small improvement in percentage terms. But for professionals relying on AI-generated analysis, 18 percent fewer hallucinations can mean the difference between a clean deliverable and an embarrassing correction.

Notably, OpenAI also says the new model produces presentations with stronger, more varied aesthetics. It makes better use of image generation tools too. So your AI-assisted pitch deck might actually look less like a template and more like something a human designer touched.

This One Isn’t for Everyone

Here’s where things get a little frustrating if you’re on a standard plan. GPT-5.4 is not available to Free, Go, or Plus subscribers. It lives in Thinking mode and the Pro model picker, which means it’s aimed at enterprise customers and developers using OpenAI’s Codex app.

For API customers, OpenAI is pricing one million input tokens at $2.50. That’s up from $1.75 with GPT-5.2. So yes, GPT-5.4 is the company’s most token-efficient reasoning model yet, but each of those tokens costs more than before.

Why OpenAI Is Betting on Productivity

The professional pivot makes a lot of sense when you look at OpenAI’s financials. The company generates about $25 billion in annualized revenue according to reporting from The Information. But it’s nowhere near profitable, and it carries more than $1.4 trillion in data center commitments.

Meanwhile, Microsoft reportedly added Anthropic’s Claude to Copilot 365 last September because Claude outperformed OpenAI’s models at tasks like generating spreadsheets and presentations. That stung, and GPT-5.4 looks like a direct response.

Professionals who rely on AI for real work represent a sustainable revenue stream in a way that casual chatbot users simply don’t. If OpenAI can own that space, it has a genuine path toward building a business that doesn’t depend entirely on investor funding.

GPT-5.4 isn’t the flashiest announcement, and it won’t matter to most people checking the weather or asking for recipe ideas. But for the developer debugging code at midnight or the analyst building financial models, this release is exactly what they’ve been waiting for.