ASTRA COMMAND CENTER
Astra, tracked as a product—not a rumor.
Every Astra fact that matters at launch, dated and sourced.
Access, pricing, limits, safety posture, and the published research record — each one marked confirmed, reported, or still unknown, with the source it came from.
WHERE THINGS STAND
Astra is out. These are the details confirmed so far.
Every detail below is tied to current OpenAI documentation. If an item is unconfirmed, the official sources do not answer it yet.
THE CURRENT RECORD
What is known, what is claimed, and what is still missing.
Each item shows its status and source so you can see what to rely on. Where OpenAI has said nothing, the record says so.
Availability
Availability
The access questions that change first when a public entry point exists.
Public availability
ConfirmedReleased September 3, 2026; access is rolling out in stages.
Release is confirmed; availability is not yet universal across every plan or account.
Public / private status
ConfirmedReleased September 3, 2026 with access expanding across ChatGPT, Work, Codex, API, GitHub Copilot, and Microsoft Foundry routes; each route still has its own eligibility rules.
ChatGPT availability
ConfirmedGPT-6 Pro, powered by GPT-6 Astra, is rolling out in ChatGPT for Pro $100, Pro $200, Business, and Enterprise. Plus plans include Astra in ChatGPT Work and Codex as it rolls out; availability can differ between Chat, Work, and Codex.
The official Help Center describes a gradual rollout and does not establish that every plan, product, workspace, or region has access today.
API availability
ConfirmedThe OpenAI API documents gpt-6-astra; access is rolling out through Trusted Access and eligible accounts rather than being universal by default.
Exact model identifier
Confirmedgpt-6-astra
Public product name
ConfirmedGPT-6 Astra
Pricing
Confirmed$10 / 1M input tokens; $1 / 1M cached input tokens; $12.50 / 1M cache-write tokens; $50 / 1M output tokens.
The current model page also lists higher multipliers for prompts above 272K input tokens and separate processing discounts; subscription access is a separate question.
Access tiers
ConfirmedTrusted Access for eligible enterprises; ChatGPT, Work, Codex, GitHub Copilot, Microsoft Foundry, and API access each depend on the current product, plan, account, client, workspace, or provider terms. Free API access is not supported.
Waitlist
UnknownNo Astra-specific public waitlist is documented in the current official sources.
The HowToUseAstra email form is an independent update list, not an OpenAI access path.
Geographic restrictions
UnknownRegional availability is not fully specified on the current model page.
EU data residency has a documented Fast mode limitation; check the current OpenAI availability and data-residency documentation for deployment decisions.
Capability and limits
Capability and limits
Fields that must come from product documentation, not research headlines.
Rate limits and quotas
ConfirmedTiered limits are published: Tier 1 500 RPM / 500K TPM; Tier 2 5K RPM / 1M TPM; Tier 3 5K RPM / 2M TPM; Tier 4 10K RPM / 4M TPM; Tier 5 15K RPM / 40M TPM.
The model page also publishes batch-queue limits; account eligibility and endpoint-specific behavior still apply.
Context window and max output
Confirmed1,050,000-token context window; 128,000-token maximum output.
These are model/API limits, not a promise that every ChatGPT plan exposes the same working limits.
Reasoning controls
ConfirmedReasoning effort supports low, medium, high, xhigh, and max; none is not supported.
Fast mode is unavailable for EU data residency, and the model guide recommends Responses API for new tool-using work.
Tools, browsing, code, files, and multimodality
ConfirmedText and image input, text output, and documented support for browsing, computer use, software engineering, science, and professional work; audio and video are not supported.
Responses tools listed in the model docs include web search, file search, computer use, code interpreter, image generation, hosted shell, MCP, and tool search.
Cybersecurity capability threshold
ConfirmedOpenAI says Astra is its first model to reach the Critical cybersecurity capability level.
This is OpenAI’s assessment, not an independent certification or unrestricted-use guarantee.
Safety and operations
Safety and operations
Deployment, monitoring, and integration facts with visible attribution.
SDK and integration support
ConfirmedUse the Responses API with the official OpenAI SDKs; the quickstart shows JavaScript, Python, C#, Java, Go, and Ruby examples for gpt-6-astra.
Chat Completions remains supported, but tool calling requires Responses API.
Safety and operational posture
ConfirmedOpenAI says Astra is its first model to reach the Critical cybersecurity capability level and describes stronger safeguards, prompt-injection robustness, and asynchronous misalignment monitoring.
The system card also describes reduced monitorability under adversarial conditions; safety claims remain deployment- and task-dependent.
Future training status
ConfirmedFuture training runs are separate from the released GPT-6 Astra product record.
Do not use future-version training context as evidence of today’s availability or capabilities.
Evidence and freshness
Evidence and freshness
The source trail and the difference between an official record and an independent test.
Confirmed published capability record
ConfirmedOpenAI documents state-of-the-art performance in computer use, browsing, software engineering, science, and professional work; customer stories add workflow-specific reports.
These are official documentation and customer-story claims, not independent HowToUseAstra test results.
Latest official announcement
ConfirmedSafety overview: GPT-6 Astra — September 3, 2026
Independent product findings
UnknownNo independent public Astra task battery has been completed by HowToUseAstra yet.
The benchmark lab is ready for a controlled run when an eligible test account and reproducible access are available.
Published cyber evaluation record
ConfirmedOpenAI reports 100% on ExploitBench and two zero-days during its evaluation; the system card provides the launch methodology and conditions.
These are vendor-reported evaluation results and do not describe default production behavior or authorize offensive use.
Computer-use report
ConfirmedOpenAI’s model guidance documents computer use as a supported capability; no independent public task battery is recorded yet.
This replaces the reported human-parity headline with a narrower, product-documented claim.
System card
ConfirmedPublished September 3, 2026 as a 117-page GPT-6 Astra system card.
The card is the primary source for tested limits, safeguards, and deployment conditions.
Last verified
ConfirmedSeptember 3, 2026
HOW WE CHECK UPDATES
What we check before changing this page.
We look for an official source, a working route where relevant, and independent evidence before calling a claim confirmed.
- Official sources.Check a launch or access change against current OpenAI documentation and, where relevant, a working official entry point.
- Exact details.Keep the model ID, endpoint, supported parameters, pricing, limits, and eligibility exactly as published.
- Independent tests.Keep vendor claims separate from results we have run ourselves, with the conditions and limitations visible.
- Reader guidance.Make the account, region, deployment, and safety limits clear before you rely on a claim.
OFFICIAL SOURCES
Why the current state is cautious.
OpenAI’s launch record names GPT-6 Astra, documents multiple access routes, publishes API and pricing details, and links the 117-page system card. That is a useful starting point, not a substitute for checking your account, region, and deployment controls.