How to Use GPT-6 Astra—and Why It Isn’t Showing

Pick the route that matches your task

For a conversation, start in ChatGPT and inspect the model options your account actually shows. For a task involving files, research, or a longer deliverable, inspect Work. For a repository or a task involving developer tools, inspect Codex. For a product you are building, check API access in the intended organization and project.

These routes share model technology but have separate access controls and usage rules. A successful Astra task in Codex does not establish that the same account can select GPT-6 Pro in Chat. Equally, an API error does not prove that your ChatGPT subscription is broken. Write down the route before troubleshooting; it narrows the possible causes.

• OpenAI’s product-by-product availability

Using GPT-6 Pro in ChatGPT

Sign in to the account with the eligible plan, confirm the active workspace, and inspect the model picker for GPT-6 Pro. The current rollout covers Pro $100, Pro $200, Business, and Enterprise. Enterprise access also depends on workspace model permissions. Availability can reach different products at different times.

Plus users should check Work or Codex for Astra as the rollout reaches them. The current help page does not list Plus for GPT-6 Pro in ordinary Chat. Before paying for an upgrade, confirm that the intended plan includes the product you need and read its model allowance. An upgrade is a poor diagnostic experiment when the underlying issue is the wrong workspace.

• GPT-6 Pro plan eligibility

Using Astra in Work or Codex

Open the intended task surface, select Astra if your account exposes it, and give it a bounded job with a deliverable. For a creative website task, attach the brand references, specify the pages and viewport sizes, and say what counts as complete. The model can use only the files and tools that the application makes available.

For Codex CLI, OpenAI specifies version 0.153.0 or newer for Astra. Check the installed version with codex --version and update through your normal installation method if necessary. OpenAI also asks users to update the ChatGPT desktop app. After an update, verify the signed-in identity and model selection again; updating software does not grant account entitlement.

• Required versions and rollout notes
• Work and Codex usage rules

Using Astra in an API application

The documented API model ID is gpt-6-astra. Start with a small text request in the intended organization and project before adding a large file collection or complex tools. The model documentation lists Responses, Chat Completions, and Batch support; OpenAI’s latest-model guide uses Responses for getting started.

Treat access, request compatibility, and application behavior as separate checks. Astra’s supported reasoning levels start at low and include medium, high, xhigh, and max; it does not list none. Copying a configuration from another model can therefore fail even when the model is available. Inspect the actual error and the current reference rather than removing parameters at random.

Keep API credentials on the server, and share redacted errors when requesting help. A useful diagnostic includes the endpoint, model ID, client version, timestamp, project identity, and request ID when available. An API key or a full private customer brief is unnecessary for a public troubleshooting post.

• Astra API specifications
• OpenAI’s Astra integration guide

If Astra is missing, check identity before software

First inspect the signed-in email and active workspace. A personal account and a company workspace can expose different models even in the same application. Then compare your plan with the availability page for the exact surface. If the plan is eligible, check the application version and whether the rollout has reached that surface.

For a managed workspace, ask the administrator to inspect both model availability and your role’s permissions. OpenAI’s enterprise guide distinguishes Chat, Work, Codex, and API access and explains that a model setting alone does not grant access. Some initial enterprise availability is tied to the Daybreak program. A model visible to a colleague may still be unavailable to your role or workspace.

• Enterprise model access and permissions

If it disappeared, inspect usage and service state

Reaching a model allowance can temporarily change which options are available. Chat’s Pro-model allowances and Work/Codex usage are separate. Check the displayed reset time and the applicable dashboard before assuming a model has been removed from your subscription. Switching between models that share an allowance does not create fresh usage.

Record the exact error or banner before refreshing or reinstalling. If several features fail at once, consult OpenAI’s service-status page linked from its Help Center. If only one project fails, focus on that project’s access and configuration. This preserves evidence and avoids changing several variables before you know which one matters.

• Usage limits, fallbacks, and reset behavior

Make the first task small enough to verify

Once Astra is available, try an illustrative creative task: review an existing landing-page screenshot and produce a prioritized revision brief. Supply the target audience, conversion goal, and the elements that must stay. Ask for observations tied to visible evidence and a clear distinction between an observed problem and a suggested experiment.

Review that result before authorizing implementation. In an agent application, access to tools and approval settings determines what it can do next. File access, network access, and permission to publish are separate concerns. Give the task the resources it needs, then verify the final artifact in the browser or exported file rather than accepting a completion message alone.

• Codex sandbox and approval controls

Choose a useful next step when access is unavailable

If the account is eligible but the rollout has not reached it, use an available model for work that can proceed and retain the brief for later comparison. If the blocker is a workspace permission, send the administrator the product name and observed symptom. If an API request fails, reduce it to a reproducible example before escalating.

For a task whose only deliverable is an image, a dedicated image tool may already meet the requirement. Ciyo offers GPT Image 2 independently of whether Astra appears in your OpenAI account. Choose that route for image production on its own merits; it does not unlock GPT-6 access.

• Ciyo’s GPT Image 2 tool

Continue reading

• Understand Astra’s plans and limits
• Understand the Astra model names