The catalog

Every routable model is one row: price per 1M input/output tokens, context window, a capability grade, and which modalities/params it supports. Browse it interactively at /models (filter by price, context, provider, modality), compare 2 or more side by side at /compare, or see it ranked by capability at /leaderboard. Programmatically: GET /v1/models.

Capability grades

Each model carries a 0–5 grade for reasoning, coding, and overall quality. This is our own heuristic, used to rank the leaderboard and to pick a model under model:"auto" — it is not an independently measured benchmark (no Artificial Analysis / LMArena / SWE-bench score is shown or implied). Every model page labels it as such.

Context window sizing

Under model:"auto", your actual estimated prompt size is compared against each candidate's real context_limit — a model that can't fit your prompt is never selected. This isn't a coarse "long/short" bucket; it's sized against your specific request every time.

Freshness

The catalog is synced on a schedule and every sync diffs prices against the previous one, flagging any model whose price moved meaningfully since last sync. GET /health reports catalog size and how stale the last sync is — a useful signal if you're pinning exact prices in your own cost projections.