GPT-6 Astra Is Here: What OpenAI’s ‘AGI Era’ Model Actually Means for You

OpenAI just rolled out GPT-6 Astra, the new model behind ChatGPT and Codex, and it’s built to do a lot more than chat. If you’re on a paid plan, Astra is landing in your account over the next few days, and it can now click through real websites, fill out spreadsheets, and carry a multi-step task through to the end almost on its own.

Here’s the short version before we get into the details:

• Release date: Rollout began September 3-4, 2026, starting with select enterprise customers.
• Who gets it: ChatGPT Plus, Pro, Business, and Enterprise users, plus the OpenAI API, Microsoft Azure, and AWS Bedrock.
• Price: Included in existing ChatGPT subscriptions; API access runs $10 per million input tokens and $50 per million output tokens in Standard mode.
• What’s new: A much stronger computer-use agent that browses, fills forms, and finishes workflows nearly twice as fast as the previous model.
• The catch: OpenAI itself classifies Astra as “Critical” risk for cybersecurity, its highest safety tier yet.

What Is GPT-6 Astra, Exactly?

Astra is OpenAI’s newest flagship model, and the company is pitching it as “the world’s most intelligent and aligned model” it has shipped. Under the hood, it’s the result of OpenAI’s largest training run to date, run on the company’s Stargate infrastructure with an earlier model helping supervise the training process.

In practice, that translates into a model that’s noticeably better at software engineering, scientific reasoning, and cybersecurity analysis than GPT-5.6 Sol, the version most ChatGPT users have been using this year. The headline change, though, is what Astra can do with a computer once you hand it the keyboard.

When You’ll Actually See It in Your Account

Don’t expect Astra to show up the moment you open ChatGPT. OpenAI is staging this rollout, starting with a small group of enterprise customers through its gated Daybreak program before opening it up more broadly. Access is off by default even for eligible accounts, so enterprise admins have to flip it on.

Here’s roughly how the access tiers break down as the rollout expands:

It’s part of the same wave of AI-heavy updates hitting your devices this month. Google leaned hard on Gemini’s item-memory tricks in the September 2026 Android feature drop, and Apple is doing its own AI overhaul with the rebuilt Siri coming in iOS 27.

The Computer-Use Upgrade That Actually Matters

Most model upgrades are about smarter answers. Astra’s biggest jump is about smarter hands. OpenAI built it to act as a computer-use agent, meaning it can navigate a browser, a spreadsheet, or a desktop app the way a person would, clicking buttons and filling fields instead of relying on a purpose-built integration.

That’s a real shift for anyone doing repetitive digital busywork, like updating a CRM, building a website draft, or reconciling a spreadsheet against a report. Astra finishes the same benchmark tasks in less time and with fewer wrong turns than its predecessor:

That last row matters more than the speed gains. It’s a test of how often the model wanders outside the task it was actually given, and going from roughly 1-in-2 to essentially never is the kind of change that makes handing over real accounts feel less reckless.

OpenAI Calls This the “AGI Era.” Should You Believe It?

OpenAI president Greg Brockman told reporters, “welcome to the AGI era,” while also admitting the line for what counts as AGI is still “gray” and “fuzzy.” That’s a fair summary of the whole launch: genuinely big capability jump, wrapped in marketing language that outruns what the benchmarks actually prove.

The more useful number, if you ask me, is the safety one. According to 9to5Mac, Astra is the first OpenAI model to hit the “Critical” tier on the company’s own cybersecurity risk framework, meaning it can find unknown software vulnerabilities and chain them into working exploits without much human help. OpenAI says it has added monitoring and stricter jailbreak defenses to compensate, but chief scientist Jakub Pachocki was blunt about the tension, warning that progress in intelligence doesn’t guarantee progress in alignment.

Worth noting too: launch materials skipped GDPval, OpenAI’s own benchmark for real-world economic work, despite Astra’s launch being framed around exactly that kind of impact. Read that omission however you like.

Frequently Asked Questions

Is GPT-6 Astra available on the free ChatGPT plan?

Not at launch. OpenAI is rolling Astra out first to Plus, Pro, Business, and Enterprise accounts, so free users will likely see it later, if at all, once demand on OpenAI’s infrastructure eases.

How much does GPT-6 Astra cost through the API?

Standard mode runs $10 per million input tokens and $50 per million output tokens. A Fast Mode option doubles the speed at roughly double the price.

Is GPT-6 Astra actually AGI?

No, and even OpenAI’s own leadership hedges on that claim. It’s a major jump in benchmark scores and computer-use reliability, not a declaration that the model can do any job a person can.

Do I need to turn anything on to get GPT-6 Astra?

If you’re on a personal Plus or Pro plan, it should arrive automatically as the rollout reaches your account. On Business or Enterprise plans, an administrator has to enable access first, since it’s off by default.

The Bottom Line

GPT-6 Astra is the most capable ChatGPT model yet, and the computer-use gains are the part that will actually change your day-to-day workflow, not the AGI headline. I’d hold off on treating it as a fully trustworthy autopilot for sensitive accounts until a few months of real-world use back up that 0% unauthorized-scope number, but if you’re paying for Plus or Pro already, there’s no reason to skip trying it the moment it shows up.