GPT-6 Astra Lands in GitHub Copilot: the US$10 Pro Plan Does Not Get It, and It Bills by Usage, Not by Request (2026-09)

On 2026-09-04 GitHub announced that OpenAI’s GPT-6 Astra is generally available in GitHub Copilot.

Two sentences matter more than the headline, and neither is in it: the US$10 Copilot Pro plan does not get this model, and it is billed by usage at provider list pricing rather than counted per request. Everything below comes from the official changelog and GitHub’s own pricing pages.

1. Which plans get it

The announcement lists Copilot Pro+, Max, Business and Enterprise.

Not on the list: Copilot Free and Copilot Pro — the US$10 tier. That differs from the Gemini 3.8 Flash rollout a couple of days earlier, which did include Pro. So if you are on Pro and cannot find GPT-6 Astra in the model picker, you are not waiting your turn — your tier does not have it.

Where you can pick it, if you do have it, is broad: Visual Studio Code, Visual Studio, the Copilot CLI, the Copilot coding agent, the Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode and Eclipse. GitHub also says the rollout is gradual, so check back if it has not appeared yet.

2. The line to notice: this one bills by usage

One sentence in the changelog is easy to skim past: the model is billed at provider list pricing under usage-based billing.

In plain terms: the more you use it, the larger the bill. There is no “so many goes a month and then it stops” — you are charged on the volume of content going in and coming out.

GitHub’s own model price list is specific (per million units):

So the same model jumps to a second price band once the input gets long — input doubles, output rises too. That matches the rule we recorded when checking OpenAI’s own price page.

3. Turning that into “how many goes does my plan cover?”

GitHub’s usage unit is the AI Credit, officially defined as 1 credit = US$0.01. Included monthly allowances, from the official plans page:

A worked example (our own arithmetic from the official list prices, not a figure GitHub publishes): a reasonably large task sending 100K units in and producing 10K units out costs US$1.0 plus US$0.5 — US$1.5, or 150 credits.

• Pro+ at 7,000 credits → about 46 tasks of that size.
• Business at 1,900 per user → about 12.
• Enterprise at 3,900 per user → about 26.

Your real numbers move with content length and cache hits, but the order of magnitude holds: this is a model where a few runs a day is worth watching the bill for, not one for casual questions.

4. Administrators: it is on by default

The changelog is explicit. Under the model policy in Copilot settings, new models are enabled automatically unless an administrator has turned off the global default or explicitly disables this model.

Combine that with the previous section and you have the practical impact for organisations: a usage-billed model appears in your members’ pickers by default. If your credits are pooled across the organisation, that allowance can be spent without anyone deciding to spend it.

Three administrator actions, all in Copilot settings:

1. Check whether default model enablement is currently on or off.
2. Decide whether GPT-6 Astra should be open to everyone or limited to specific teams.
3. Tell members that this model is not billed the way the ones they usually pick are.

5. Is it worth using?

GitHub describes it as built for long-horizon, autonomous coding work, and says that in internal testing it plans and validates as it goes, batches diagnosis with verification, and independently confirms its results before declaring a task done — reaching the end of long tasks in fewer steps than earlier OpenAI models.

That places its value in large, complete pieces of work, not one-line questions. The test is simple:

• A long task you want carried through with fewer round trips → worth the ~150 credits.
• A small edit, a syntax question, one line of documentation → use a cheaper model and keep the bill off it.

Announcement and pricing checked on 2026-09-05 against the official GitHub changelog, the Copilot models and pricing reference and the Copilot plans page. Prices and plans can change at any time; the official pages are authoritative. Related checks: GPT-6 Astra’s 272K double billing, the Copilot 9/1 credit cliff, Gemini 3.8 Flash in Copilot and our GitHub Copilot tool page.