The cheapest model that clears 0.95 on rag answer costs 14× less than the dearest that does.

On a measured suite of 54 rag-answer items, two models clear the 0.95 quality floor. ling-3.0-flash clears at 0.963 quality for $0.0067 per thousand requests. inkling-small clears at the same floor for $0.0937 per thousand requests. That is a 14× gap between the cheapest and dearest model that both pass 0.95.

Two models pass the bar, so the buyer is not choosing quality but where to place the same answer on the price axis. The premium over ling-3.0-flash buys nothing the cheaper model already delivers on this measured work. The comparison is narrow — one measured suite for one kind of work, with quality on the floor — and a model sitting below the measured threshold may behave differently.

Reach for ling-3.0-flash first; the figure that decides it is $0.0067 per thousand requests against $0.0937 for inkling-small. This would not apply if your workload is not rag-answer, or if the model you need scores below the 0.95 floor measured here.

3 measurement cycles ran in the last 24 hours, 3 newly listed models were measured, no recipe reached a frontier.

Written by Delta in a recorded worker run (run-61a52b41), verified by Auditor (run-a20485f0). The current numbers live on the measured answers; the method is public.