OpenAI GPT-6 Astra rollout: who can access Work, Codex and the API

OpenAI GPT-6 Astra rollout: the current status

As of 14 September, OpenAI says GPT-6 Astra has launched in ChatGPT Work, Codex and the API, while Work is still gradually reaching eligible accounts. Both parts of that sentence matter. The 14 September business announcement is the clearest current product-status statement: Astra is "now available" across those three routes. The current Help Center page, however, says that an eligible account may not yet have Work as the rollout continues.

This is therefore not an invite-only model in the broad sense, nor is it a universal account-level entitlement. The practical result depends on the route a person is using: ChatGPT Work, Codex or the API; the plan or organisation; the account and workspace; and, for Work, the point that account has reached in the staged rollout. A product announcement can establish launch without proving that every eligible user can already select the product.

The chronology explains the apparent tension. On 3 September, OpenAI described a phased launch beginning with a limited set of organisations and extending over the following days to paid ChatGPT plans and the API. CNBC independently reported that initial phased approach. The later OpenAI statement supersedes that initial timetable for product availability, but the Help Center’s account-level rollout language remains the necessary caveat.

A plan name is not the same as an access surface

The most common source of confusion is treating Chat, Work and Codex as the same product surface. OpenAI describes Chat as the regular conversational experience, Work as an agent for longer multi-step work and finished deliverables, and Codex as the dedicated software-development experience. Astra may be available on one of these surfaces without creating the same model option or the same usage rules on another.

OpenAI’s current documentation says Plus includes GPT-6 Astra in Work and Codex. It separately says that GPT-6 Pro, which is powered by Astra, is available in regular Chat for eligible Pro, Business and Enterprise plans; Enterprise access also depends on workspace model-access permissions. That distinction is material for someone comparing a Plus subscription with a managed company account. It is inaccurate to say that Plus brings GPT-6 Pro to regular Chat.

Why ChatGPT Work can still be absent

OpenAI’s Help Center says Work is gradually rolling out to eligible accounts over the coming days. If Work is absent, the documentation says the account may not have access yet. Eligibility is not a published promise of immediate visibility, and OpenAI has not disclosed the share of eligible accounts that had received Work by this article’s source-check time.

For an individual, the sensible reading is narrow. Confirm that the intended account and, where relevant, workspace are selected; check the plan’s stated surface; and use the current app where desktop access is involved. OpenAI says Astra requires Codex CLI version 0.153.0 or later and advises users to check for a current desktop-app update if a model is missing. It also states that updating can reveal available features but does not change plan, workspace or rollout eligibility.

Work and Codex share usage, but they are not interchangeable

Work is available on the web and mobile for eligible paid plans, and can be available in the desktop app when the plan and workspace include it. Codex remains a distinct desktop experience for development work; it is not selectable as a web or mobile experience, although supported desktop Codex chats may be accessed remotely from the mobile app. This is a product-design distinction, not evidence that every device or account has the same Astra access.

OpenAI says Work and Codex draw from a shared allowance. Plus and Business Standard include limited Astra use, while the documentation says Pro plans and Business Premium seats can use their full existing allowance. Usage changes with the task, input and output size, reasoning setting and Fast mode; the Help Center calls its message figures estimates rather than fixed limits. Credits can pay for eligible additional usage, but OpenAI explicitly says they do not add model access.

Business access is an administrator-controlled deployment

For a business customer, the question is not only which subscription exists. OpenAI says Enterprise administrators can enable Astra under their applicable rate card and agreement, and that Enterprise access was off by default at launch. The company’s Help Center adds that Enterprise model availability can depend on workspace model-access permissions. These are reasons an employee can be on an otherwise eligible plan yet not have the model in a particular workspace.

OpenAI describes additional controls for organisations, including role-based access and separate settings for Work Cloud, Work Local, Codex Local, browser use and network access where supported. Administrators can set a starting model and other defaults for Work and Codex, while Chat has a separate starting default. These are deployment and permission controls, not a way to grant unavailable features or turn off OpenAI’s platform safety measures.

The API route: model name, token price and limits

Developers can identify the API model as gpt-6-astra. OpenAI says the model is available through its API. That confirms a public product route, but it does not establish identical availability for every organisation, endpoint, rate card, capacity condition or data-retention arrangement. Zero Data Retention, for example, is described as available to eligible API customers on supported endpoints and subject to approval.

The published Standard API rate is $10 per million input tokens and $50 per million output tokens. This is the only price that can be stated cleanly from the current official material: it is usage-priced API context, not the cost of a ChatGPT plan, a flat monthly charge, an India price or a complete bill. OpenAI says separate cache-read and cache-write rates apply, and Fast mode costs twice the Standard price.

Critical cybersecurity is OpenAI’s risk classification

OpenAI calls Astra its first model to reach the Critical cybersecurity capability level under its Preparedness Framework. In the company’s description, the classification concerns what the model could do with appropriate tools and access. It should be read as OpenAI’s capability and risk assessment, not as an independent certification, a consumer-plan label or a catalogue of features granted to an ordinary user.

That difference changes how the rollout should be interpreted. A normal Work, Codex or API route does not entitle a customer to every capability examined in OpenAI’s evaluation material. OpenAI’s own documentation presents the classification alongside a stronger safety and deployment approach. It does not support the claim that paid access means unrestricted cybersecurity functionality, and this article does not describe operational cyber techniques.

Advanced cybersecurity workflows remain restricted

OpenAI says access to Astra’s most advanced cybersecurity workflows is more limited than ordinary deployment. Its pre-launch deployment note says advanced work was initially available to a small group of alpha testers, with Daybreak Blue intended to expand defensive use. It also states that a cited set of Astra results reflected Daybreak Blue access rather than the default production configuration. Those qualifications are central, not footnotes.

OpenAI says it uses layered protections including model safety training, refusals, system classifiers, automated review and monitoring for tool-using external deployment. It warns that these checks can slow, pause or stop legitimate work. In ChatGPT and Codex, a paused task may ask for review before continuing; OpenAI says an API task will stop. These are the company’s descriptions of its controls, not a guarantee that safeguards eliminate risk or interruption.

What OpenAI has disclosed about limits and uncertainty

The safety discussion is not only a list of claimed protections. OpenAI says Astra’s monitorability declined relative to GPT-5.6 Sol in adversarial evaluations and that it is investigating the implications. Its system card also cautions that evaluation settings and their scope limit what observed results can establish. The disclosure is relevant because it prevents a simple claim that monitoring is comprehensive or that the model is reliable in every setting.

Reddy News treats the launch, access routes, plan treatment, price and country lists here as official claims from OpenAI. CNBC’s 3 September report independently corroborates the original phased rollout and the early restricted programme context. OpenAI’s performance, alignment and safety results remain the company’s own claims unless independently validated; this explainer does not convert them into promises about any individual task or deployment.

India support is real, but it is not an Astra guarantee

India appears on OpenAI’s current supported-country lists for ChatGPT and for the OpenAI API. That is useful confirmation that the underlying services are supported from India. It is not, however, an Astra-specific availability table. The same country-list material warns that access outside supported territories can result in an account being blocked or suspended.

The defensible conclusion for an Indian reader is therefore limited: ChatGPT and API services are supported in India, while Astra access remains dependent on the relevant product route, plan or API organisation, staged Work rollout, account standing, workspace permission, allowance and safety controls. OpenAI has not publicly specified in the material reviewed an Astra-specific India launch date, Indian price, quota, restriction or Daybreak eligibility path. Do not infer one from the country lists.

What to verify before changing a plan or workflow

Start with the intended destination, rather than a model name alone. A reader seeking longer work should check whether Work is visible and whether the account is still in the staged rollout. A developer should assess API organisation access and token-based costs. A managed employee should ask the workspace owner about model-access permissions and the relevant role. Someone who needs regular Chat should separately confirm whether the plan offers GPT-6 Pro there.

Several details remain unknown in public documentation: the percentage of eligible accounts with Work enabled, Astra-specific India timing and billing, individual enterprise rate cards, regional capacity and the criteria or timetable for restricted Daybreak access. Those gaps make a precise promise impossible. The most accurate status remains straightforward: GPT-6 Astra is launched across Work, Codex and the API, but access is still shaped by surface, rollout, plan, workspace and safety controls.