What can replace my model?
Compare alternatives with the model you use. See the gain and what you give up.
At least 20% cheaper for a typical coding-agent task. A model more than 20 points less intelligent than yours is left out; a smaller drop is named.
Your alternatives
How the answer is chosen
It is arithmetic, not an AI: the same model and wish always give the same answer. A candidate has to be clearly better at what you asked for — at least 20% cheaper or faster, more open, or with a longer context. Intelligence is what it must keep: within 10 points of yours it is the best fit, 10 to 20 points lower a good alternative, and more than 20 points lower it is not offered at all. A model whose intelligence nobody has measured is only worth a look. Within a verdict a clearly bigger gain goes first (a fresh number counts in full, an old one for less); among gains of a similar size, the smaller drop in intelligence goes first. A vendor's “0” is not a price, so such a model is never called cheaper.
Up to five options, best verdict first. Within a verdict a clearly bigger saving leads; among savings of a similar size, the model that keeps more intelligence goes first.
The full method