GPT-6 Astra Is Live for Limited Orgs — Not GA Yet

As of September 4, 2026, GPT-6 Astra is rolling out to a limited set of organizations. It is not generally available. That is the status that matters — Help Center and ChatGPT release notes are still the product tape, even after OpenAI's own X account claimed a wider wave for some paid seats.

OpenAI's ChatGPT release notes still say access is rolling out to a limited set of organizations, that Astra is not yet generally available, and that broader availability is planned over the coming days. The live Help Center articles for GPT-5.6 and GPT-6 Pro and ChatGPT Work and Codex describe a gradual rollout to Plus, Pro, Business, and Enterprise plans and warn that Astra may not be immediately available on your plan. That gradual / not-GA frame is the primary read on this page.

Separately, OpenAI posted on X that Astra is now available to all Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex, is live in the API, and may take a few days for Plus and Business. Treat that as what the announcement account said — linked below — not as a merge into "Pro is live for everyone on every surface." Do not revive the stopped "Pro / Enterprise / Business Premium live" headline on this slug.

This post keeps the same URL as our August write-up so Social's link and the search listing stay intact. The August framing — "cannot call, cannot price, launch delayed" — is stale. The math story from early August is still useful background. The access story is staged, contested across OpenAI surfaces, and still not GA.

What changed since the August "delayed launch" read

In early August, OpenAI published research claims about Astra on hard math and theoretical computer science problems, then followed with a Preparedness Framework note that put Astra at Critical on the cybersecurity track. At that point there was no public endpoint and no public price. That was accurate then.

It is not accurate now. OpenAI has published a live model page for gpt-6-astra, listed token pricing, and started a limited-organization / gradual paid-plan rollout across ChatGPT and API surfaces. The Critical cyber designation still shapes how access ships — gated first, broader later — but it no longer means "held forever" or "you cannot plan around a price."

If you landed here from an older social post or a search snippet that still says delayed, treat that line as outdated as of this update. If you landed here from a "Pro is live" rewrite, treat that merge as wrong: Help Center still says limited / gradual / may not be immediate.

Availability: Help Center first, OpenAI's X claim second

Prefer Help Center and release notes for product status. Cite the OpenAI X post for what the announcement account claimed. Where those disagree, say so — do not invent "full Pro GA."

• Help Center — ChatGPT release notes (help.openai.com/en/articles/6825453): introducing GPT-6 Astra; access rolling out to a limited set of organizations; not yet generally available; broader availability planned over the coming days.
• Help Center — GPT-5.6 and GPT-6 Pro in ChatGPT (help.openai.com/en/articles/20001354): Astra "is rolling out gradually to Plus, Pro, Business and Enterprise plans, and may not be immediately available in your plan." In Chat, GPT-6 Pro (powered by Astra) is rolling out for Pro $100, Pro $200, Business, and Enterprise; Enterprise access still depends on workspace model-access permissions. Pro includes Astra in Chat, Work, and Codex as it rolls out; Plus includes Astra in Work and Codex as it rolls out. Availability can differ across those products during rollout.
• Help Center — ChatGPT Work and Codex (help.openai.com/en/articles/20001275): same gradual Plus / Pro / Business / Enterprise wording and the same "may not be immediately available" caveat. Once Astra is available to your account, Pro $100, Pro $200, and Business Premium seats can use their full existing Work/Codex allowance for Astra; Plus and Business Standard seats include limited Astra usage, with optional credits afterward. Buying credits does not buy early access during rollout. "Business Premium" here is an allowance rule for when the model reaches you — not proof it is already live for every Premium seat.
• API model page (developers.openai.com/api/docs/models/gpt-6-astra): live model id gpt-6-astra, public pricing table, 1,050,000-token context, 128,000 max output tokens, April 30, 2026 knowledge cutoff, and tools including computer use on the Responses API. The older Trusted Access "rolling out today for enterprises…" lead is not on the live model markdown checked for this update.
• OpenAI on X — separate announcement claim (status/2095968413646737608): GPT-6 Astra "is now available to all Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex." "It's also live in the API." "It might take a few days to roll out to our Plus and Business users." That is what the tweet says. It is not what the release notes lead with, and it is not a license to title this page "Pro is live."

So the honest read: limited orgs / not GA / gradual on Help Center and release notes; OpenAI's X account claims a Pro / Enterprise / Business Premium Work+Codex+API wave with Plus and Business still rolling; ChatGPT Chat for GPT-6 Pro is plan-gated and may lag Work/Codex; Free and Go stay out of the current Pro tables. Check your own model picker and workspace settings.

Older copies of this page talked about "Trusted Access" vs "Daybreak" program labels on the Astra rollout. Those labels are gone from the live Astra Help Center articles and the live API model page we checked for this update, so they are dropped here. (Daybreak / Trusted Access for Cyber still exists as a separate Help Center program for authorized cyber workflows — that is not the Astra availability label anymore.)

API pricing as listed on the model page

As verified on the live API model page on September 4, 2026, standard text-token pricing for GPT-6 Astra is:

• Input: $10.00 per 1M tokens
• Cached input: $1.00 per 1M tokens
• Cache writes: $12.50 per 1M tokens
• Output: $50.00 per 1M tokens

Source: OpenAI API docs — GPT-6 Astra. That page also notes prompts over 272K input tokens are billed at 2x input/cache and 1.5x output for the full request, and that Batch and Flex are 50% of standard rates while Fast mode is 2x.

Help Center ChatGPT plan pages describe included subscription allowances and optional credits for Astra usage once your account has the model. They are not a substitute for the API per-token table above. If you are estimating API spend, use the API page. If you are estimating ChatGPT Work or Codex burn, use the Help Center plan rules for your seat.

The August math claim is still OpenAI's count

Back in early August, OpenAI said Astra produced "10 advances in mathematics and theoretical computer science" on problems that had gone more than ten years without recorded progress. That claim was produced inside OpenAI, on a model OpenAI controlled, described in a document OpenAI wrote. Third parties still cannot independently rerun those exact problems on a public harness the way they can rerun a published SWE-bench chart.

An "advance" is also a wider word than "solved." It can mean a full proof, a special case, a tighter bound, a counterexample, or a construction that unsticks later work. The count is useful as a signal about long-horizon formal reasoning. It is not a shopping metric. It does not tell you whether Astra will finish your migration script without inventing an API.

The Critical cyber designation that followed those math posts still matters for product shape. A model rated that high on cybersecurity work is the kind of release that starts with limited organizations, workspace toggles, and usage review. The September rollout matches that pattern: limited orgs and gradual paid access first, Free and Go left out of the current Help Center Pro tables.

What to do this week if you actually need Astra

If you are on Pro, Enterprise, Business Premium, or any other named paid plan, check whether gpt-6-astra appears in your API project and whether GPT-6 Pro / Astra appears in ChatGPT Chat, Work, and Codex for your workspace. Do not assume a plan label — or an OpenAI tweet — equals instant access on every surface during a staged rollout. Enterprise still depends on workspace model-access permissions.

If you are on Plus or Business Standard, Help Center still puts you in the gradual / may-not-be-immediate bucket for Work and Codex, with limited Astra allowance once the model reaches you. OpenAI's X post says Plus and Business may take a few days. Plan against what your account actually shows.

If you are writing production API code, pin expectations to the live model page: model id gpt-6-astra, $10 / $50 standard text rates, huge context, and tools that include computer use, code interpreter, web search, and MCP on the Responses API surface listed there.

For day-to-day work that cannot wait on rollout luck, ChatGPT on GPT-5.6 Sol, Codex, Claude, and Gemini remain the callable defaults. Astra is no longer a pure research rumor. It is also not a finished GA launch. Limited orgs / not GA on the Help Center tape. OpenAI's X claim is wider — and separate. Check the picker, not the hype.