Since 4 September 2026 OpenAI has been rolling out its new top model: GPT-6 Astra — the thing many have been searching for as “ChatGPT 6” for months. The headlines talk about the “AGI era”. If you are wondering what of that actually reaches you: that is exactly what this article is for. The short answer: for your subscription, less changes than the headlines suggest — for the way AI operates your computer, more.
I will walk through the announcement the way I did with the Claude update a week ago: what is new, what it costs, who gets it — and what you can ignore.
What is GPT-6 Astra?
GPT-6 Astra is the successor to GPT-5.6 Sol, which only arrived in July. OpenAI opened it to selected partners on 3 September and, since 4 September, to all paying ChatGPT users — in waves “over the coming days”. There are two tiers: Astra for Plus plans and Astra Pro for Pro, Business and Enterprise. In company accounts it is switched off at launch and has to be enabled by the admin.
The context window is around one million tokens, the knowledge cutoff April 2026. On the developer API the model is called gpt-6-astra and also runs on Microsoft Azure and Amazon Bedrock.
What does GPT-6 cost?
For you as a user: nothing extra. Plus subscribers get Astra with their plan; Pro, Business and Enterprise get Astra Pro. No extra credits, no new tier — unlike Anthropic's Fable, which on the Pro plan only runs on usage credits.
On the developer API, where billing runs by consumption:
| Price per M tokens (API) | GPT-6 Astra | Claude Fable 5.1 |
|---|---|---|
| Input | $10 | $10 |
| Output | $50 | $50 |
| Cache read | $1 | $0.25 |
| Fast mode | double the price, up to 2.5× faster | — |
| Context window | ~1M tokens | 1M tokens |
The base prices are identical to Anthropic's top model. The difference is in re-reading: on long-running tasks where the AI reads the same material over and over, Claude is four times cheaper. What that means day to day is explained here: Claude Fable 5.1: stronger than Opus 5 and cheaper to run.
What does GPT-6 Astra do better?
The number that matters is not the maths record. It is the one for operating a computer: in the OSWorld 2.0 test, where the model completes real tasks on screen — opening programs, clicking, editing files — Astra completes 72.6 percent of tasks, its predecessor 65.7. And it needs around 47 percent less time to do so. At finding the right button in crowded interfaces it reaches 92.7 percent instead of 76.9.
Translated: the AI that operates your computer is becoming usable. Not perfect — almost three in ten tasks still go wrong. But the jump from “toy” to “just do it” is exactly the one that matters for solopreneurs: filling in forms, moving data from a portal into a spreadsheet, recurring click sequences.
Add reliability: in OpenAI's own task collection Astra solves 88 percent on the first attempt, its predecessor 56. Less fixing, less supervising.
The catch: the safety tier
GPT-6 Astra is the first OpenAI model to reach the company's highest internal risk level for cybersecurity. Without safeguards it could exploit previously unknown vulnerabilities in tests. That is why you get a restricted version that refuses certain cybersecurity requests; the full version only goes to vetted security firms.
You know the pattern from Anthropic: Fable 5.1 for everyone, Mythos 5.1 only for vetted organizations. For your daily work it is irrelevant — unless you work in IT security. For everyone else it means: the models have reached a point where the makers no longer ship them without a brake.
GPT-6 or Claude — do you have to switch?
No. And for the same reason you would not have had to switch to Claude a week ago if you are on ChatGPT: the difference between the top models is smaller than the difference between a tool that lives in your workflows and one you are just getting to know.
Both cost the same. Both have a million tokens of context. Both get better every few weeks. What differs is where your projects, your instructions and your routines live. If you built them in Claude, switching costs more time than GPT-6 will ever save — and vice versa.
If you are still at the fundamental question of which of the two fits your business: I wrote that down after half a year on both sides — Claude vs. ChatGPT: which AI fits your business? GPT-6 has not changed that.
What you should do now
- If you have ChatGPT Plus: nothing. Astra turns up in your model menu on its own. Pick it for long, multi-step tasks; for everyday texts the faster model remains the better choice.
- If you work with Claude: also nothing. Your reference point is Claude on a subscription, not GPT-6 on the API.
- If you have recurring click work: make a note of “computer use”. That is the capability becoming usable right now — at OpenAI and at Anthropic. Collect the three workflows you would most like to hand off. The tools for them are arriving now.
Which model is ahead changes every few weeks. Whether your workflows benefit is up to you. That is exactly what we talk about in my free community “KI — aber richtig” — the templates and workflows from my videos live there too. And if you want company instead of doing it alone: the AI Business Community offers monthly live calls and workshops for exactly that.
Frequently asked questions
What does GPT-6 Astra cost?
Nothing extra on a ChatGPT plan: Plus users get Astra, Pro, Business and Enterprise users get Astra Pro. On the developer API it costs $10 per million input tokens, $50 for output and $1 for cache reads; fast mode costs double.
Is GPT-6 included in ChatGPT Plus?
Yes. Since 4 September 2026 OpenAI has been rolling GPT-6 Astra out in waves to all paying ChatGPT users, Plus included. The stronger Astra Pro tier is only available on Pro, Business and Enterprise; in company accounts the admin has to enable the model first.
What is the difference between GPT-6 Astra and Astra Pro?
Astra is the version for Plus plans, Astra Pro the one for Pro, Business and Enterprise. Both are based on the same model; Astra Pro is the tier for more demanding tasks. Both ship in the restricted version that refuses certain cybersecurity requests.
Is GPT-6 better than Claude?
In the published tests, GPT-6 Astra and Claude Fable 5.1 sit at the same price level with equally large context windows; Astra shines at operating computers, Claude at cheaper continuous operation. For most self-employed people the deciding factor is not the test score but where projects, instructions and routines already live.
Conclusion
GPT-6 Astra is a big update with no pressure to act: for Plus users it arrives on its own, for Claude users it changes nothing, and the real progress — AI that operates the computer — still needs a few months before it reaches your daily work. Until then, collect the workflows you want to hand off. Whoever has done that is productive within an hour at the next update. Whoever only reads the headlines starts from zero every time.
