Grok 4.7's low sticker price costs more than Astra in practice
Counting output tokens, Grok 4.7 burns about 81,000 tokens per task for a $2.73 bill, while Astra reaches the same score for $0.82.
grok 4.7 lists $2 in and $6 out, but that rate card is a lie once the harness starts thinking, because the model now dumps about 81,000 output tokens per intelligence-index task against 27,000 for gpt-6 astra at max so the cheap sticker turns into a $2.73 bill while astra does the same 46-point score for $0.82. push grok 4.7 to xhigh and the score does not move it stays 46 while the cost climbs to $3.74, whereas astra at maximum effort reaches 53 for $3.26, which means you are paying more tokens for a weaker ceiling. on terminal-bench the same pattern shows up in the total: grok 4.7 burned through $3,683 across 324 of 330 trials, astra max sat near $3.3k, This problem needs to be solved
This post quotes another post: x.com/vvtentt101/status/2102061362042085865