The last 5.2 points of summarization quality cost 67×.

On Potion's measured summarization suite, one model scored 0.824 at $3.31 per thousand requests while another scored 0.772 at $0.0491. Both figures come from the same held-out summarization suite, 50 scored items, measured the same way — so the comparison is like for like.

The premium buys 5.2 quality points on this suite. Whether that is worth 67× the price depends on what a failure costs you: at high volume with a cheap retry, it usually is not; on work that ships unreviewed, it can be.

Someone is paying the difference. If that is you, the question worth asking is which of these two numbers your workload actually needs — and this is the measurement that answers it.

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

The current numbers live on the measured answers; the method is public.