How to Use GPT-6 Astra: API Key, Playground, Codex, and Cursor Setup

Quick Answer

To use GPT-6 Astra you have two doors: the OpenAI API (create a key at platform.openai.com and call the model ID gpt-6-astra) or ChatGPT, where it appears as "GPT-6 Pro" on Pro, Business, and Enterprise plans (Plus access is currently limited to Codex and Work — more on that below). This is a how-to, not a what-is: for specs and background, see What Is GPT-6 Astra. Here we cover getting access, the exact code, and the pricing traps to avoid before your first request.

Step 1: Get API Access

1. Sign in at platform.openai.com and open API keys under your organization settings.
2. Create a new secret key and copy it — you won't see it again.
3. Add billing (a prepaid credit balance works; the API has no free credits for Astra).
4. Note the base URL: https://api.openai.com/v1.

The model string matters. The API model ID is gpt-6-astra — if you copy an older snippet that says model="gpt-6", it predates the launch naming and will error or resolve to something unintended. Astra shipped September 3, 2026; anything written before that date is suspect.

Step 2: Make Your First Call

Astra speaks the standard OpenAI SDK protocol, so a minimal Python example looks like this:

from openai import OpenAI

client = OpenAI(api_key="YOUR_KEY")  # base_url defaults to https://api.openai.com/v1

resp = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[
        {"role": "system", "content": "You are a precise coding assistant."},
        {"role": "user", "content": "Refactor this function and explain the diff."},
    ],
)
print(resp.choices[0].message.content)
print(resp.usage)  # watch input/output/cache tokens — this is your bill

Two habits worth starting on call #1: log the usage field on every response, and put anything stable (system prompts, tool definitions) at the front of messages so it can hit the cache. The full API access guide covers keys, org setup, and gateway configuration in more depth.

Step 3: ChatGPT Access (No Code Path)

If you don't want the API, Astra is available in ChatGPT — but which plan gets you what is messy:

Quota numbers are community-reported, not officially documented — treat them as directional. If you were hoping for free access, the honest answer is in Using GPT-6 Astra for Free.

Step 4: Playground and Codex

• Playground: after picking gpt-6-astra from the model dropdown, use the Playground to test system prompts before committing them to code — cached-prefix structure you validate here is what saves money later.
• Codex: the notable Astra-specific behavior is context preservation and retrieval when the window fills. In long sessions, Codex no longer just truncates; it preserves and retrieves earlier context. Practical implication: you can run longer autonomous tasks without the mid-session amnesia that forced manual re-briefing on older models. Full walkthrough: GPT-6 Astra in Codex and Cursor.

Step 5: Cursor and IDE Setup

In Cursor (or any OpenAI-compatible tool): open model settings, add a custom model named gpt-6-astra, and point it at your API endpoint. If your tool supports a custom base URL, that's also where a gateway drops in — same key, same model string, different endpoint. The setup details, including config-file snippets, live in the Codex/Cursor setup article — we won't duplicate them here.

Cost Mechanics to Know Before Your First Call

Astra is $10 per million input tokens and $50 per million output. Three things move that number:

1. Caching is 10x cheaper on input. Cache reads are $1.00/M vs $10.00/M fresh. Structure prompts so the stable prefix is cacheable — this is the single biggest lever.
2. Long context has a surcharge tier. Requests beyond roughly 272K input tokens are reported to rebill at double rate. The ~1.05M window is real, but the top of it costs 2x per token — dump the whole repo in only when you mean it.
3. Computer use is token-hungry. Astra is the strongest computer-use model (~72.6% OSWorld), but screenshot → action loops burn input on every step. Budget accordingly.

For the full math, see GPT-6 Astra pricing explained, and for cutting the bill down, cost optimization tactics.

FAQ

Q: What model ID do I use for GPT-6 Astra in the API? gpt-6-astra. If a snippet says model="gpt-6", it predates the launch naming (September 3, 2026) and won't route to Astra.

Q: Can I use GPT-6 Astra in ChatGPT on the Plus plan? Currently only through Codex and Work — Plus subscribers have largely lost direct chat access to it, which has become an active complaint on OpenAI's forum. Direct chat access effectively starts at the Pro tiers.

Q: Does Codex work differently with Astra? Yes — Codex gains context preservation and retrieval when the context window fills, so long agent sessions can keep working from earlier context instead of losing it.

Q: How do I reduce the cost of my first experiments? Cache your system prompt (reads drop from $10 to $1 per million), stay under ~272K input tokens to avoid the reported 2x tier, and route routine subtasks to a cheaper model. A gateway makes the routing part trivial: TeamoRouter gives you gpt-6-astra and cheaper fallbacks behind one key, pay-as-you-go.

Q: Do I need a US credit card for API access? Not necessarily — OpenAI takes prepaid credits, and gateways like TeamoRouter accept Alipay and WeChat Pay with the same model available at base_url https://api.teamorouter.com/v1.

Summary

Using GPT-6 Astra comes down to: an API key from platform.openai.com (or a gateway), the exact model string gpt-6-astra, and awareness of the two cost tiers — cache reads at $1.00/M and the reported 2x billing past ~272K input. ChatGPT access exists but is plan-gated, with Plus currently limited to Codex/Work.

Get Started

TeamoRouter — one API key, base_url https://api.teamorouter.com/v1, GPT-6 Astra alongside Claude and cheaper fallbacks, pay-as-you-go with Alipay/WeChat support.