Prices in USD per 1K tokens, cacheRead being the row's cache-read
rate or 0.1x its input price, and longContext (its own input, output
and cacheRead over fromTokens) on a model priced by prompt length;
undefined for an empty, unknown or unpriced id.
A Claude model's prices for a host that admits a call before it is made, as
openAIModelPricingprices an OpenAI call: the host strips anyanthropic:prefix first, as it does for OpenAI. Resolution is by an exact catalog id, else the id with its dated suffix removed, else the catalog row that is the id plus a dated suffix (a retired snapshot only by its exact id); an id the catalog does not name this way (claude-opus-4,claude) isundefined, never the provider's prefix guess, so a spend guard refuses what it cannot price instead of pricing it as another model.