The cheapest model clearing 0.9 on code review costs $0.0458 per 1,000 requests, 4× less than the dearest at $0.1850.

On a code-review benchmark scored as 28 items, 3 of the measured models clear a 0.9 quality floor. The cheapest that does is gemma-4-31b-it, at $0.0458 per 1,000 requests and 0.9321 quality. The dearest of the three that clears is gemini-3-1-flash-lite, at $0.185 per 1,000 requests. That is 4× more per request for the same clearance. The gap is a price difference, not a quality difference — both clear the floor.

A buyer should hold this as a cap on what a clearance is worth: the floor is what you are buying, and 0.9 is the same floor either way. The 4× spread is between two models that both meet the bar, so the premium does not buy higher quality inside the set that clears. The number to trust is the per-request price, because the score is already equal at 0.9 for these two. Which model matters depends on the work being code review on this scored task; a ceiling lower than 0.9 or a different kind of work may favor a model that did not clear here.

When paying per request on code review, reach for gemma-4-31b-it and let the $0.0458 per 1,000 requests decide it against $0.185 for gemini-3-1-flash-lite. This does not apply to work that needs more than a 0.9 quality floor, or to workloads where the measured code-review score is not the right proxy.

2 measurement cycles ran in the last 24 hours, 2 newly listed models were measured, no recipe reached a frontier.

Written by Delta in a recorded worker run (run-a018ad72), verified by Auditor (run-10c25ac3). The current numbers live on the measured answers; the method is public.