GPT-6 Astra Is Live, But Not for Subscribers Yet — and the 272K Double-Billing Line (2026)

OpenAI shipped its new flagship, GPT-6 Astra, in September 2026. Most coverage led with benchmark claims. We went straight to OpenAI’s official developer model page instead, and found two things that matter more to ordinary users than any score: you probably cannot use it yet, and there is a 272K threshold buried in the price table.

(Every number below comes from reading OpenAI’s official model page on 2026-09-05. Where the official page is silent, we say so.)

First things first: who actually has access?

The official model page says access is currently rolling out to enterprises through OpenAI’s Trusted Access Program, with broader API and subscription access described as coming soon.

In plain terms:

• ChatGPT Free: no access.
• Plus / Pro / Business / Enterprise: the official page says “coming soon”, not “available”. Open your model picker and you may well not see it.
• General API developers: same queue. Enterprises first.

Some outlets have reported that Plus, Pro, Business, Enterprise and AWS will be switched on within days. That sentence does not appear on the official model page, so we label it as press reporting rather than confirmed fact. The fastest way to check your own access is the model picker in ChatGPT, not a headline.

The specs: a 1.05M context window, April 2026 cutoff

Here is what the official page lists:

A million-token context is roughly a full documentation set. But note the text-only output: no image generation, no audio. The official page also lists what is not supported right now — Realtime, Assistants, fine-tuning, and the audio and image generation endpoints. Supported endpoints are Chat Completions, Responses and Batch.

The real catch: that 272K line

The headline API pricing is simple enough:

But directly underneath, the official page adds a condition that changes the maths: requests exceeding 272K input tokens incur 2x input and cache rates, and 1.5x output pricing.

That matters because it sits on the same page as the 1.05M context window. Plenty of people see a million-token window and plan to throw an entire corpus at it. In practice:

• Input at or under 272K: input US$10, output US$50.
• Input over 272K: input becomes US$20, cached input US$2, output US$75.

Put bluntly, “the context window goes to 1.05M” and “using 1.05M costs the same rate” are two different claims. If you genuinely need to feed long documents, work out whether you can stay under 272K, or chunk and then synthesise. The gap here is not a few percent — your input rate doubles.

What this means if you just use ChatGPT

If you only use the web app or the mobile app:

1. Nothing changes for now. Subscription access has not broadly rolled out.
2. When it does, your monthly bill does not change either — subscriptions are not token-metered, and the 272K double-billing rule is API-only.
3. The thing to actually watch is the 2026-04-30 knowledge cutoff. Ask about anything after April and you need it to search the web; do not take an unsourced answer at face value.

What this means if you build on the API

Write the 272K line into your cost model, especially for RAG and long-document summarisation, where context tends to balloon. Older models did not have this step change, so carrying over an old formula will under-estimate your bill.

Should you switch now?

The official positioning is “our most capable model, built for the hardest end-to-end work” — complex reasoning, coding, computer use, research, document creation. That is also a way of saying it is not the value pick for everyday chat. A rough rule:

• Everyday Q&A, email, summaries → your current model is fine. Wait for pricing or subscription access.
• Long-horizon agent work, cross-file code changes → worth evaluating, but price in the 272K step first.
• You need images, audio, or fine-tuning → not supported here. Don’t spend time evaluating it.

All figures checked on 2026-09-05 against the official OpenAI model documentation. Pricing and availability can change at any time; the official page is authoritative. For a full plan comparison, see our ChatGPT tool page, and for the paid-tier breakdown, is ChatGPT Go worth it.