GPT-6 Astra is now available to all ChatGPT Plus, Pro, Business, and Enterprise users
GPT-6 Astra expands beyond its limited preview to all paid ChatGPT tiers, the OpenAI API, Azure, and AWS Bedrock: with faster computer use and new Codex note-keeping.
GPT-6 Astra made its debut on 3 September 2026, initially available only to a small set of enterprise organisations in OpenAI’s Daybreak cybersecurity programme. By the evening of 4 September 2026, OpenAI had expanded access to all paying ChatGPT subscribers on Plus, Pro, Business, and Enterprise plans, alongside the OpenAI API, Microsoft Azure, and AWS Bedrock. If you have a paid ChatGPT subscription (excluding the $8/month Go plan), Astra is now available to you.
Who gets access, and how
Access follows your existing subscription tier. Plus, Pro, Business, and Enterprise users all get Astra within their current allowances, with the option to purchase additional usage credits on top. Pro, Business, and Enterprise subscribers also get access to GPT-6 Astra Pro, the higher-capability variant.
There is one important exception for enterprise workspaces: access is turned off by default. If you are a workspace admin, you need to enable Astra manually in your Workspace settings before it appears for your users. This is a deliberate choice on OpenAI’s part, not an oversight, and it gives IT and security teams a chance to review the model’s capabilities before it goes live across an organisation.
Users on the $8/month Go plan are not included and will not get access to Astra or GPT-5.6.
For developers, Astra is available via the OpenAI API as gpt-6-astra, and through both Microsoft Azure and Amazon Bedrock. This matters because it means organisations that have already built procurement and compliance relationships with those cloud providers can adopt Astra without routing everything through OpenAI’s own platform. Zero Data Retention is supported for eligible API customers.
API pricing is $10 per million input tokens and $50 per million output tokens, with cached input at $1. Batch and Flex modes are available at half the standard rate. The model carries a 1,050,000-token context window and a 128K maximum output, with a knowledge cutoff of 30 April 2026. Note that prompts exceeding 272K input tokens are charged at 2× the standard input rate.
What’s new in this rollout: Codex note-keeping
One of the more practically useful additions in the Astra rollout is a change to how Codex handles long sessions. Previously, when a coding session grew long enough to fill the context window, Codex would compress earlier turns into a summary and discard the originals. That works well enough for straightforward tasks, but it loses detail in complex debugging sessions: why a particular fix failed, which tests ran, which requirement was added three hours ago.
Astra introduces cross-context-window note-keeping. Rather than discarding earlier turns, Codex now stores notes across context windows and keeps earlier messages searchable. If Astra needs to retrieve a requirement or a test result from much earlier in the session, it can find it even if the note didn’t explicitly capture that detail.
This feature is experimental at launch and needs to be enabled in your Codex config.toml. OpenAI says it will become the default for Astra in the coming weeks. If you work on long refactors or complex debugging sessions, it is worth enabling now.
Computer use is significantly faster
Astra brings roughly a 2× improvement in computer-use speed compared to GPT-5.6 Sol. In latency simulations on OSWorld 2.0, Astra completed tasks in around 40 minutes, compared to roughly 75 minutes for GPT-5.6, while also improving accuracy from 65.7% to 72.6%. On the Mind2Web benchmark, the combined effect of Astra’s efficiency and the updated Codex harness translates to approximately 1.9× faster task completion.
In practical terms, this means tasks like filling out online forms, updating CRM records, organising calendars, manipulating spreadsheets, or working within engineering tools such as KiCad and FreeCAD are not just more accurate but meaningfully quicker. For teams exploring AI-assisted workflows that go beyond drafting and question answering, the speed improvement changes what’s realistic to delegate.
The safety context you should know
GPT-6 Astra is the first OpenAI model to reach the “Critical” level of cybersecurity capability under OpenAI’s Preparedness Framework, meaning it can, in principle, identify previously unknown security vulnerabilities and develop exploits autonomously. The publicly available version refuses advanced offensive security work. More capable, less restricted access to those features is being routed separately to vetted defenders through OpenAI’s Daybreak Blue programme.
OpenAI also built a new alignment evaluation specifically informed by its Hugging Face incident in July 2026, which tested whether a model facing a difficult task would exceed its authorised scope. GPT-5.6 Sol did so 48% of the time in testing without production safeguards in place. Astra did so in 0% of cases. That is a meaningful shift in model behaviour, not just a policy layer on top.
Within ChatGPT and Codex, the model may pause and ask you to review an action before continuing. In the API, tasks can stop when a safety check is triggered. OpenAI is clear that monitoring is an additional layer and not a substitute for alignment training.
A note on the rollout friction
The launch was not entirely smooth. Some Pro subscribers expected to be first in line based on their subscription tier and were frustrated when enterprise customers in the Daybreak programme got access earlier. Sam Altman acknowledged this publicly and confirmed Pro subscribers would go first in the broader rollout. OpenAI also offered subscribers one banked reset for each day their account remained without access.
It is worth knowing this context because it explains why the rollout happened the way it did, and because it signals that OpenAI is now managing access to its most capable models with more deliberate sequencing, not simply flipping a switch for everyone at once.
What this means for you
If you are a ChatGPT Plus, Pro, Business, or Enterprise subscriber (on a plan above Go), Astra is available to you now. The upgrade is included in your existing plan.
If you manage a ChatGPT Enterprise workspace, your first step is to enable Astra in Workspace settings. Nothing changes for your users until you do.
If you are building with the OpenAI API, or deploying through Azure or AWS Bedrock, gpt-6-astra is available today. Review the pricing and context window details before migrating, particularly the token-rate changes above 272K input tokens.
If you use Codex for extended coding sessions, enable cross-context-window note-keeping in your config.toml now rather than waiting for it to become the default.