llmrelay
// gpt

GPT-6 Astra

OpenAI’s current flagship. API, ChatGPT paid plans, half list here.

GPT-6 Astra is OpenAI’s current flagship, released 3 September 2026. The model id is gpt-6-astra. OpenAI’s own page puts it on ChatGPT Plus, Pro, Business and Enterprise, and on the API, Azure and Bedrock. Free and Go do not include it. There is no separate free Astra API quota — usage sits inside those paid allowances, or you buy credits.

On the API, OpenAI’s short-context list is $10 per million input tokens and $50 per million output, with cached input at $1 and cache writes at $12.50. Long-context requests double that. Fast mode is 2× standard. We sell the same id at half list: $5.00 / $25.00. Point any OpenAI-compatible client at https://api.llmrelay.dev/v1 and set model to gpt-6-astra.

OpenAI positions Astra for computer use, browsing, software engineering, documents and research. ChatGPT Sites can build and host a page or a game from a prompt. That is computer use plus tools, not a native 3D generator. Reddit demos that one-shot a 3D game are the model driving Blender, Three.js or a browser — useful, and easy to overclaim.

Official list
llmrelay
Input / M tokens
$10.00
$5.00
Output / M tokens
$50.00
$25.00
Context window
—
1,050,000 tokens
Max output
—
128,000 tokens

What it costs you per month

Real-world budget scenarios. Numbers are simple sums — official list price vs llmrelay's 50% off tier.

Usage scenario
Official
llmrelay
Light coding (1M in / 200K out per month)
$20.00
$10.00save $10.00
Heavy Cursor / Cline user (50M in / 5M out per month)
$750.00
$375.00save $375.00
Production RAG (500M in / 20M out per month)
$6000.00
$3000.00save $3000.00

What it's good at

Best for

Pick GPT-6 Astra when

  • +You already live in ChatGPT, Codex or the OpenAI API and want the current flagship without paying list. Same id, half the invoice.
  • +The job is a long agent loop: fill a form, update a CRM, review a repo, keep going when the context window fills. That is the work OpenAI actually trained it for.
  • +You want a second opinion next to Claude. One llmrelay key covers gpt-6-astra and claude-opus-5, so the comparison is a model-id change, not a new vendor.
  • +You saw the 3D-game clips and want to reproduce the workflow in an API you control, not only inside ChatGPT.

Choose something else when

  • !You need a native image or video model. Astra can drive tools that make those; it is not Sora and it is not GPT Image 2.
  • !The work is cheap classification or short routing. GPT-5.6 Luna or Claude Haiku 4.5 will do that at a fraction of $5/M input.
  • !You are cost-sensitive on everyday coding. Claude Sonnet 5 is $1.00/M input here. Astra is the escalation, not the default.
  • !You only have a ChatGPT Free or Go account and you expected a free API. OpenAI does not include Astra on those plans, and neither do we as a free tier.

Questions people ask about GPT-6 Astra

How can I use GPT-6 Astra?

Three official doors: ChatGPT Plus/Pro/Business/Enterprise, the OpenAI API as gpt-6-astra, and Azure or Bedrock. Free and Go are out. On llmrelay you use the same model id against https://api.llmrelay.dev/v1 with one prepaid key. Cursor, Codex CLI, or any OpenAI-compatible SDK works if it lets you set a base URL.

Does GPT-6 Astra have a free quota?

Not as a standalone free API. OpenAI folds Astra into existing paid ChatGPT allowances and lets you buy extra credits. We do not run a free Astra pool either — $10 minimum top-up, billed per token at $5.00/$25.00.

Can GPT-6 Astra make 3D games?

It can drive the tools that make them. OpenAI’s own demos are computer use: write code, open a browser, operate Blender or Three.js. That is not a native 3D modality like a video model. If the clip you saw was a one-shot playable prototype, the model wrote and ran software — it did not emit a mesh from a single image endpoint.

How much does GPT-6 Astra cost?

OpenAI list is $10.00 per million input tokens and $50.00 per million output on short context, $20/$75 when the prompt is long, cached input $1.00. On llmrelay it is $5.00 and $25.00 — half the short-context list, one rate regardless of prompt length. Fast mode at OpenAI is 2×; we do not currently expose a separate fast tier.

Is GPT-6 Astra worth it over Claude Opus 5?

On llmrelay Astra is $5.00/$25.00 and Opus 5 is $2.50/$12.50, so Astra costs twice as much per token. Use Astra when you need OpenAI computer use, Codex, or the ChatGPT Sites workflow. Use Opus 5 as the default coding model. Run the same prompt on both — one key covers them — and keep the cheaper one when the output is good enough.

What is the context window on GPT-6 Astra?

1,050,000 tokens input, with up to 128,000 output tokens. Knowledge cutoff on OpenAI’s model page is 30 April 2026.

Try GPT-6 Astra at half the price

Free to create an account, no subscription, $10 minimum top-up — enough to run this model against a real task and judge quality yourself.

Get API key →

Compare GPT-6 Astra against the alternatives

The comparisons this model appears in, the models nearest it on price, and the full rate card.