GPT-6 ASTRA · PRICE AND LIMITS
GPT-6 Astra: the million-token window has a price step in the middle
OpenAI released Astra on 3 September 2026 at $10 per million input tokens and $50 per million output. The context window is 1,050,000 tokens — but a prompt over 272,000 input tokens is billed at double the input rate and 1.5× the output rate, for the whole request. Most coverage quotes the window and the headline price and stops between them.
Direct answer
What GPT-6 Astra costs, in the order that matters
Standard pricing is $10 per million input tokens and $50 per million output. Batch and Flex run at half that. Fast mode is double. Cache writes are billed at 1.25× the uncached input rate. And above 272,000 input tokens the whole request is repriced: 2× input and cache, 1.5× output — so the last third of the advertised window costs materially more per token than the first two thirds.
Comparison basis
Sources
Price, context window, the long-context threshold and the cyber capability level come from OpenAI's own model and pricing pages. VUST has no relationship with OpenAI and does not run this model.
Unless a separate method is stated above, this page does not claim a head-to-head product-quality benchmark.
GPT-6 Astra, as OpenAI publishes it
Released
Published value
3 September 2026
The part worth noticing
Rolling out to ChatGPT paid tiers, the API, Azure and Bedrock
Where it bites
Availability differs by surface
Family
Published value
GPT-6 Astra
The part worth noticing
OpenAI model and pricing pages, checked 12 September 2026, do not describe a separate Astra Pro
Where it bites
Read these figures as Astra figures only
Price
Published value
$10/M input · $50/M output
The part worth noticing
Batch and Flex at 50%, Fast mode at 200%
Where it bites
Mode choice moves cost 4× end to end
Context window
Published value
1,050,000 tokens
The part worth noticing
Above 272,000 input: 2× input, 1.5× output, whole request
Where it bites
Long-document work, not chat
Max output
Published value
128,000 tokens per request
The part worth noticing
Output is the expensive side at $50/M
Where it bites
A single long generation dominates the bill
Cache writes
Published value
1.25× the uncached input rate
The part worth noticing
Cache reads are the saving; writes are a surcharge
Where it bites
Short-lived caches can cost more than they save
Cyber capability
Published value
First model at the Critical level of OpenAI's Preparedness Framework
The part worth noticing
OpenAI's own classification, not a benchmark score
Where it bites
Access controls, not throughput
The same published price, and one of them runs in Telegram today
Anthropic publishes Claude Fable 5.1 at $10 per million input and $50 per million output — the identical headline price to GPT-6 Astra, which makes it the only like-for-like comparison on cost rather than on benchmarks. Claude Opus 5.5 sits at $4/$20, 60% less. VUST does not run GPT-6 Astra. It does run Claude Fable 5.1, selectable by name in @vustbot, so if the question behind your search is "what does a frontier model at this price actually do for my work", that is the one you can answer today rather than read about.
The model in short
OpenAI released GPT-6 Astra on 3 September 2026 at $10 per million input tokens and $50 per million output, with a 1,050,000-token context window and up to 128,000 output tokens per request. Prompts above 272,000 input tokens are billed at 2× input and 1.5× output for the entire request; Batch and Flex run at half the standard rate, Fast mode at double, and cache writes at 1.25× the uncached input rate. It is the first model OpenAI classifies at the Critical cybersecurity level of its Preparedness Framework.Open Claude Fable 5.1 in @vustbot
Frequently asked questions
How much does GPT-6 Astra cost?
$10 per million input tokens and $50 per million output at standard rates. Batch and Flex are half; Fast mode is double; cache writes carry a 1.25× surcharge on the uncached input rate. The number people miss is the long-context one below.
What is the 272K threshold?
The window is 1,050,000 tokens, but a request whose input exceeds 272,000 tokens is repriced at 2× input and cache and 1.5× output — for the full request, not the excess. So a 300,000-token prompt does not cost slightly more than a 270,000-token one; it costs roughly twice as much on input. OpenAI documents one exemption: Codex usage.
Can I use GPT-6 Astra in VUST?
No. VUST does not run it. If you want a frontier model at the same published price in Telegram, @vustbot runs Claude Fable 5.1 at $10/$50, selected by name.
GPT-6 Astra vs Astra Pro: what is actually established?
The OpenAI model and pricing pages this page cites document GPT-6 Astra. Checked 12 September 2026, they do not describe a separate Astra Pro, so nothing on them lets you give Pro the price, the context window or the limits shown here, and nothing on them supports a capability comparison between the two. That is a limit of these sources, not a verdict on whether Pro exists elsewhere: if you meet a figure or a capability claim for Pro, check which page it came from. No model named Astra is selectable in VUST today. @vustbot runs Claude Fable 5.1 by name; the models page lists the rest.
What does the Critical cybersecurity level mean?
It is OpenAI's own classification under its Preparedness Framework, not a benchmark score: the level at which a model can, with tools and access, find previously unknown security flaws and develop exploits without a person guiding each step. Astra is the first model OpenAI has placed there. It describes capability and the safeguards around it, not availability or quality for ordinary work.
Is GPT-6 Astra better than the model I use now?
OpenAI reports state-of-the-art results on computer use, browsing, software engineering, cybersecurity, science and professional work. Whether that changes your result depends on your task, and price is the part you can compare without a benchmark: at $10/$50 it is two and a half times Claude Opus 5.5 and level with the Anthropic flagship of the same class.