OpenAI announced GPT-6 Astra on September 3, 2026. GitHub followed with its Copilot availability announcement on September 4. This analysis was published and its sources checked on September 10.
The useful migration question is which route fits your work: a coding application, an API integration, or an existing Copilot workflow. These routes should be evaluated separately.
Where can you use GPT-6 Astra?
| Route | What to verify before starting |
|---|---|
| Codex | Check the model selector, active account and workspace policy in the client you use. |
| OpenAI API | Use the documented model ID gpt-6-astra; verify project access and billing. |
| GitHub Copilot | Check the eligible plan, rollout status and administrator model policy. |
OpenAI’s launch announcement covers its product rollout. GitHub’s announcement lists Pro+, Max, Business and Enterprise and says rollout is gradual. Its availability list includes the Copilot app, CLI, coding agent and supported editors.
If the model is absent, first identify the exact product and signed-in account. Reinstalling an editor will not resolve a missing plan entitlement or an organization policy. Record these details before escalating an access problem.
What does the API cost?
The OpenAI model reference lists these standard USD rates per million tokens, checked September 10:
| Token category | Listed rate |
|---|---|
| Uncached input | $10 |
| Cached input | $1 |
| Cache writes | $12.50 |
| Output | $50 |
Requests above 272,000 input tokens have higher rates for the full request: twice the input/cache rates and 1.5 times the output rate. Batch, Flex and Fast also have different multipliers; consult the live reference for the mode you select.
These are API token rates. They are not a monthly subscription quote or a forecast for a complete coding task. For planning, keep input, cache reads, cache writes and output in separate columns, then add retries, execution infrastructure and human review. GitHub says Astra uses provider list pricing under its usage-based billing; check the Copilot billing view for your actual allowance.
Should you replace your current coding model?
Our recommendation is to start with a task your present setup struggles to finish: a reproducible bug, a change spanning several files, or a migration with explicit acceptance tests. Keep the repository revision, tools, permissions and success criteria constant.
Measure accepted changes, review time, failed attempts and total billed usage. A longer answer or more autonomous activity is not an accepted result. Do not change the model and its permissions at the same time: that makes it harder to explain the outcome.
For a routine edit that your existing setup completes reliably, retain it as the baseline. Move more work only when the new route earns its place on the tasks you actually maintain. Our cost-per-accepted-task worksheet helps structure that comparison.
What this release does not establish
This is a review of provider documentation, not an authenticated Astra benchmark or a claim that every account has access. We have not verified Astra inside MonkeyCode.
Compare Claude Fable 5.1 and Mythos 5.1 if your question is model selection. For the latest DeepSeek release and the separate hosted free-use option, use the maintained DeepSeek guide.
Frequently asked questions
What is the GPT-6 Astra API model name?
Use gpt-6-astra for the OpenAI API. Confirm that the API project has access and that your integration supports the requested endpoint and options.
Why is GPT-6 Astra missing from my Copilot model picker?
GitHub lists Pro+, Max, Business and Enterprise and describes a gradual rollout. Check your plan, client and organization model policy. The September 4 announcement does not list Copilot Free or Pro.
Does a ChatGPT subscription include OpenAI API usage?
Do not treat subscription access as an API allowance. Check billing in the product you actually use: OpenAI API, Codex and GitHub Copilot have distinct access and usage rules.