GPT-6 Astra Lands in ChatGPT for Plus and Pro Users, Not Free Accounts
OpenAI released GPT-6 Astra on Thursday, and for now almost nobody can use it. The model went live first to a limited set of enterprise customers in OpenAI's early-access programme, with ChatGPT Plus, Pro, Business and Enterprise subscribers told to expect it "in the coming days." Free accounts are not on the list at all.
Which plans get Astra, and which get Astra Pro
The rollout splits into two tiers, which is easy to miss in the announcement. Plus, Pro, Business and Enterprise subscribers all get GPT-6 Astra itself. A heavier variant, GPT-6 Astra Pro, is limited to the Pro, Business and Enterprise plans — Plus does not include it.
Enterprise workspaces are a further exception. There, Astra ships switched off, and an administrator has to enable it before anyone in the workspace sees it in the model picker. If you are on a company account and the option has not appeared, that is the likely reason rather than a slow rollout.
- Free — no access announced. Free accounts stay on the current default model.
- Plus — GPT-6 Astra, within the plan's usage allowance.
- Pro, Business, Enterprise — GPT-6 Astra and GPT-6 Astra Pro.
- Enterprise — off by default until an admin turns it on.
Usage counts against your existing subscription allowance rather than being metered separately, and OpenAI is letting subscribers buy additional credits once that allowance runs out. There is no separate Astra subscription to sign up for.
What it costs developers
API pricing is where the increase is plain. Astra runs $10 per million input tokens and $50 per million output tokens, with cached input at $1 and cache writes at $12.50. That is roughly 2.5 times the previous generation's rate.
OpenAI's argument is that the per-token figure is the wrong measure, and that Astra finishes work in fewer steps — the company puts the saving at around 57 percent on some tasks despite the higher rate. That is a claim about OpenAI's own test workloads, not an independently verified result, and it will not hold evenly across every use.
The model carries a context window of about 1.05 million tokens — 922,000 in, 128,000 out — and a knowledge cutoff of 30 April 2026. Beyond ChatGPT and the API, OpenAI says it will also be reachable through AWS and Microsoft Azure.
The Codex change is the practical one
For anyone using Codex, the more useful addition is experimental: Astra can keep notes across context windows during a long session. Previous behaviour compressed everything down into a single rolling summary, which quietly lost detail on work that stretched over hours. Notes preserve the specifics instead of flattening them.
It is labelled experimental, so treat it as such. This follows a run of Codex-linked changes in the app, including the point when ChatGPT gained access to Mac iMessages for Codex and Work users.
The benchmark numbers, and one that does not match
OpenAI put Astra at 99.9 percent on ARC-AGI-3 and 74.1 percent on DeepSWE v1.1 for software engineering, plus 72.6 percent on OSWorld 2.0 for computer use, at roughly 40 minutes per task. On FrontierMath Tier 4, published accounts differ: the figure appears as both 97.6 percent and, rounded, 98 percent. The precise number is 97.6.
One result is worth reading carefully rather than as a headline. Astra scored 100 percent on ExploitBench and turned up two previously unknown zero-day vulnerabilities during evaluation, which is why it is the first OpenAI model rated "critical" under the company's own Preparedness Framework. That classification is a safety label describing capability that could be misused, not a warning about using the chatbot. OpenAI president Greg Brockman marked the launch with the line "Welcome to the AGI era" — a company framing, and one worth separating from the measurements above.
How to check whether you have it
There is no toggle that forces the rollout. Update the ChatGPT app, then open the model picker at the top of a new chat and look for GPT-6 Astra in the list; if it is absent, your account has not been switched over yet. Enterprise users should ask their workspace admin first. Free-tier users have nothing to check — the tier still runs the existing default, as it has since OpenAI dropped text limits on free accounts.
Frequently Asked Questions
Is GPT-6 Astra available on the free ChatGPT plan?
No. OpenAI listed Plus, Pro, Business and Enterprise as the plans receiving GPT-6 Astra, and free accounts were not included. Free users continue on the existing default model. OpenAI has not said whether the free tier will get Astra later.
When will GPT-6 Astra appear in my ChatGPT account?
OpenAI released it on 3 September 2026 to a limited group of enterprise customers in its early-access programme, and said paid ChatGPT plans would follow in the coming days. No exact date was given. There is no setting that makes the rollout reach you sooner.
What is the difference between GPT-6 Astra and GPT-6 Astra Pro?
Astra Pro is a heavier variant restricted to the Pro, Business and Enterprise plans. Plus subscribers get GPT-6 Astra but not Astra Pro. Both draw on the usage allowance already included with the subscription.
How much does GPT-6 Astra cost through the API?
It is $10 per million input tokens and $50 per million output tokens, with cached input at $1 and cache writes at $12.50. That is around 2.5 times the previous generation's per-token rate. OpenAI argues the cost per completed task is lower because Astra needs fewer steps.
What does it mean that GPT-6 Astra is rated critical?
"Critical" is the top classification under OpenAI's own Preparedness Framework, applied here because the model scored 100 percent on the ExploitBench cybersecurity evaluation and found two previously unknown zero-day vulnerabilities during testing. It describes capability that could be misused, not a risk to people using the chatbot normally. It is the first OpenAI model to carry the label.
