GPT-6 Astra prices, and the 272K line that reprices the whole request

Per million tokens, gpt-6-astra is $10 input and $50 output⁠^[1,2]. Cached input is $1, one tenth of the input price, and a cache write is $12.50⁠^[1,2].

Above 272,000 input tokens a second set of rates applies: $20 input, $2 cached input and $75 output⁠^[1,2]. Input and cache rates double, and output is 1.5 times⁠^[1,2].

The part that is easy to miss is that those rates apply to the whole request, not only to the tokens past the line⁠^[1]. A prompt of 273,000 tokens is billed at $20 for all 273,000⁠^[1,2]. It is not $10 for the first 272,000 and $20 for the last thousand⁠^[1,2]. Crossing the line by one token raises the price of everything before it.

Batch halves everything, to $5 and $25 on short context and $10 and $37.50 on long⁠^[1,2]. There is also a fast mode at twice the standard rate: $20 and $100 short, $40 and $150 long⁠^[1,2].

Against the GPT-5.6 line, Astra is 2.5 times gpt-5.6-sol on both input and output⁠^[2]. Against Claude, the $10 and $50 are the same numbers as Claude Fable 5.1⁠^[1,2,3].