Optimize
Every setup that fits, on one tok/s per user by $/1M map.
-
01
Optimize — every setup that fits
What the Optimize sweep enumerates: GPUs, precisions and parallelism that fit at every context stop.
4 min read · updated 2026-10-26 -
02
Reading the tok/s per user × $/1M map
How to read the log-log map: the front, the best point per GPU and precision, and what the flags mean.
4 min read · updated 2026-10-26 -
03
Goals, draggable limits and the context slider
Choose a goal, drag the caps on the map and move the context slider to see which setups survive.
4 min read · updated 2026-10-26