Every id this router serves, live from
GET /v1/models: effective prices as billed right now,
speed as MEASURED on this router (p50 time-to-first-token and
tokens/sec from live probe traffic - sparse routes say
insufficient_data, never a fake number), each model's
retention tier with the provider policy citation that earned it, and
provider jurisdiction where it is sourced (absence means unknown -
never guessed). Machine pickers fetch the compact
GET /v1/models/chooser in one call instead - the same
facts, one payload. Link to any model with # + its id;
filter and sort ride the URL too.
No quality rankings here, deliberately: this page publishes prices as billed, measured speed, and cited capability facts (context, modality, retention tier). A quality score that cannot cite its sources is marketing, and we do not print marketing as measurement.
Models
Concrete rows with their own price sheet and retention posture. A scheduled retirement shows its UTC cutover on the row; retired models leave this list at that instant. Expand a row's details for measured speed, lifecycle and the pool/class aliases that can resolve to it.
Classes
State intent instead of a model; the router picks per your account's cost sensitivity. Billed at the chosen model's own sheet.
Smart routes
Resolved per request over a live candidate ladder - the price range spans the candidates' effective sheets.
Orchestration presets
Multi-model recipes; each leg bills at that leg model's own row above.