Reading progress of “Model pickers users can trust”: 0%
Course · Shipping AI interfaces
Model pickers users can trust
A model selector is a trust device, not a settings menu — every badge it shows makes a promise, and the honest picker labels its numbers as editorial or measured.
Intermediate1 min read (computed · recorded 3)updated 2026-09-14aimodelsselectcourse
by Motif Editors
revised 2026-09-14 — First publication in the bank-4 AI-interface curriculum.
— Every chip on a model card is a promise: latency, capability, quality — label which are measured and which are editorial.
— The picker is keyboard territory: arrows, Home, End and Escape are the minimum.
— Selection should be announced, and the default should be the model you can actually stand behind.
A badge is a promise
A latency chip that reads '1.2 s' is a measurement or it is a lie. A capability badge ('code', 'writing') is an editorial judgement. The honest model picker separates the two visually and says which is which — this site prints 'latency chips are editorial, not measured' under its picker, because the alternative is a user who learns the whole surface is decoration.
Keyboard is the test
A model picker that works by mouse alone is broken. The listbox pattern answers arrows for movement, Home and End for the edges, Escape to close, and announces the selection when it changes. The default model is a product decision too: pick the one you can actually stand behind on a cold first visit.