GPT-6 Can Use Your Computer Now. Here's What Actually Changes

If you glanced at the news last week, you saw two letters and a number everywhere: GPT-6. OpenAI rolled out its most powerful model yet, called Astra, on September 3rd. And the headline everyone latched onto wasn't about it being smarter. It was that it can now sit down and use your computer the way a person does.

That's a different kind of claim than we're used to. For three years the pitch was "AI can write things and answer questions." This one is "AI can click around your software, fill out your forms, and move through your spreadsheets on its own." OpenAI's president, Greg Brockman, said it can do all of that "often at superhuman speed."

So is this the moment everything changes, or just another giant number on a benchmark chart? Let's translate it into something useful.

What actually happened

OpenAI released GPT-6 Astra, which it's calling its "most intelligent and aligned model." It came out of the company's largest training run ever, using more than 100,000 chips at its Stargate facility in Texas. If you like big scary numbers, there are plenty of them.

But strip away the specs and here's the real shift. Older AI was a great writer trapped in a text box. You typed, it typed back. GPT-6 is being sold as something that can actually operate the tools you already use. Think less "help me write an email" and more "go into the system, pull the numbers, and put them in the spreadsheet."

The rollout is gradual. It started with a small set of enterprise customers and is reaching Plus, Pro, and Enterprise ChatGPT users over the following days. The everyday consumer version is also deliberately locked down in certain areas, especially anything touching cybersecurity. So the version most people get is powerful, but on a leash. That detail matters more than it sounds, and we'll come back to it.

The part that actually matters: "computer use"

You'll hear the phrase "computer use" a lot now, so let's make it concrete.

For years, getting AI to do real work meant connecting it to your software through custom plumbing. A developer had to wire it into your systems one piece at a time. That's expensive and slow, which is why most small businesses never got past the chat window.

"Computer use" skips a lot of that plumbing. Instead of needing a special connection to each app, the AI looks at the screen, understands what it's seeing, and clicks, types, and scrolls like a person would. In theory, that means it can work inside almost any software, even the clunky old system your industry still runs on that no modern tool integrates with.

Picture the tasks that eat your afternoon. Copying invoice details from an email into your accounting software. Checking three portals for the same order status. Pulling last month's numbers out of one system and dropping them into a report. This is exactly the kind of tedious, click-heavy work "computer use" is aimed at.

If that clicks for you, you already understand why this release got so much attention. It's not that the AI got wiser. It's that it grew hands.

Why normal people should care

Here's the honest read, and it's a mix of exciting and "hold on a second."

The exciting part: the gap between "AI that talks about work" and "AI that does work" is closing fast. A year ago, an AI that could reliably run a multi-step task inside real business software was a demo that fell apart the moment you tried it yourself. The whole industry has been chasing the point where the demo matches reality. This release is a real step toward that.

The "hold on" part: a step toward is not the same as arrived. Even in OpenAI's own materials, the smartest version is being handed out carefully, in stages, with guardrails. The consumer version refuses whole categories of tasks. And "often at superhuman speed" quietly includes the word "often." When a company that badly wants you excited still hedges its own language, that's your cue to stay curious but keep your expectations on the ground.

We've made this point before in most AI releases are noise you can safely ignore. This one is worth understanding, not because you need to rush out and rebuild your business around it, but because the direction it points is the one that eventually reaches your desk.

What changes for you tomorrow morning?

Realistically? Not much, yet. And anyone telling you otherwise is selling something.

The flashy "computer use" abilities are rolling out slowly, cost more, and still need a human watching. This is not a "turn it on and go to the beach" situation. If anything, this connects straight to a lesson we keep hammering: the more an AI can do on its own, the more it needs a person checking its work, not less. A model that can click buttons in your live systems raises the stakes, not lowers them, which is exactly why set it and forget it is the biggest lie about AI. An AI that can fill out a form correctly can also fill it out wrong, fast, in a place that's hard to undo.

So no, you don't need to panic or pivot this week. But there are a few things worth watching.

What business owners should actually watch

Over the next few months, four things are worth keeping an eye on, and none of them is the launch itself.

When "computer use" reaches the tools you already pay for. The real moment for you isn't the OpenAI announcement. It's when your CRM, your booking software, or your accounting tool ships an "AI can do this for you" button built on this stuff. That's when it gets practical for a normal business.

Whether it's reliable, not just impressive. A demo that works nine times out of ten is a great demo and a terrible employee. Watch for whether these tools become boringly dependable, because boring and dependable is what you actually want running your back office.

What it costs once the hype cools. The most capable versions carry a premium right now. Prices in this space have a habit of falling, which is the whole story of the AI price war. The smart move is often to wait a beat, let the price settle, and let other people find the bugs first.

The oversight question. If you ever let an AI operate inside your real systems, who reviews what it did, and how would you catch a mistake? Figure that out before you flip the switch, not after.

Is this hype, or does it actually matter?

Both, which is the most useful answer I can give you.

It matters because the ceiling on what AI can do just went up in a direction that touches real work, not just clever writing. The idea of software that can run other software is a big deal over the next few years.

It's also hype because "the ceiling went up" and "your Tuesday is different now" are two very different sentences, and companies love to blur them. The capability is real. The finished, trustworthy, affordable, drop-in-for-your-business version is not here yet. It's coming. It's not here.

That's actually good news for you. It means you have time to get your feet under you before this becomes standard, instead of scrambling later. The businesses that will win with this aren't the ones chasing every launch. They're the ones who understood the direction early and got comfortable with the basics while everyone else was still arguing about benchmarks.

What to actually do this month

You don't need GPT-6 to start. You need reps. A simple plan:

• Make a list of your click-heavy tasks: the copy-paste, check-the-portal, move-numbers-between-systems stuff. These are the exact jobs "computer use" is coming for, so knowing yours is step one.
• Practice with what you already have. Get fluent with the AI tools in front of you now. The people who benefit most from the next release are the ones who were already comfortable with the last one.
• Pick one process to understand deeply. The clearer you are on how a task actually works, the easier it is to hand off later, and part of that clarity is knowing when not to use AI at all.

None of that requires a shiny new model. It just requires paying attention to the right things.

If you're trying to figure out where AI actually fits into your business, and which of these headlines matter for you versus which are safe to ignore, that's exactly what we help people sort out at Humanity AI.