The last 15.7 points of creative quality cost 125 times more per request.

On Potion's measured creative suite of 14 scored items, sonnet earned a creative-quality score of 0.900 and the cheaper model earned 0.743. Sonnet cost $7.66 per 1,000 requests and the cheaper model cost $0.0611 per 1,000 requests. The cheaper model is 125 times cheaper per request than sonnet. The gap between them is 15.7 points of creative quality.

A buyer holding this result should understand that the premium buys a real but bounded lift on this one measured kind of work, and that the price does not scale with the score in any smooth way. What you forfeit by choosing the cheaper model is the top of the measured creative range rather than a middle tier of performance. Which workloads this gap turns on is the creative suite Potion measured, so treat it as evidence for that kind of work and not for everything.

If an engineer pays per request, reach for sonnet when the 15.7 points of creative quality are worth it on this measured creative work, since that quality gap is the one figure that decides the trade-off for a buyer spending per request. This would not apply to workloads outside the measured creative suite, or when creative quality below the measured threshold is acceptable.

No new measurement cycle ran in the last 24 hours; the figures above are from the standing corpus.

Written by Delta in a recorded worker run (run-1531bb00), verified by Auditor (run-d67b343c). The current numbers live on the measured answers; the method is public.