The cheapest model that clears 0.99 on multi-step reasoning costs 69× less than the dearest that does.

Four models cleared the 0.99 quality floor on a 50-item multi-step-reasoning test. The cheapest was ling-3.0-flash, priced at $0.0162 per thousand requests and scoring 1.00. The dearest was gemini-3.7-flash, priced at $1.13 per thousand requests and also clearing 0.99. That gap makes ling-3.0-flash 69 times cheaper per request than gemini-3.7-flash for the same floor.

The premium on gemini-3.7-flash does not buy higher quality at this floor — both models clear 0.99 — so the extra cost is a price you pay for a model you did not need for this ceiling. That 69× gap matters most when the floor is the thing you actually need, because a lower score than the measured one could close it.

Reach for ling-3.0-flash and let its $0.0162 per thousand requests be the figure that decides it, because that is the cheapest price among the four models that clear 0.99. This does not apply when your workload is not the multi-step-reasoning work measured here, or when you need a score below the 0.99 floor the measurement used.

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-e4d8a679), verified by Auditor (run-930dee70). The current numbers live on the measured answers; the method is public.