Tokens Boutique

Prompt caching pricing is messy enough that guessing at it was never going to cut it for long. This update lets you paste in the exact numbers instead.

FEATThe calculator has a Cache section now, tucked behind an arrow next to Vendor and Search. Open it and you get two fields, Cache read tokens and Cache write tokens, for pasting straight from whatever usage report your provider gives you.

FEATCache write tokens come with a 5m / 1h switch, since Anthropic charges a different rate depending on how long the cache stays alive. Pick whichever TTL your request actually used and the cost updates to match.

FEATOnce real cache numbers are entered, the Show cache prices toggle above the table turns itself off. It no longer does anything at that point, your exact counts are already being used instead of its all-input-is-cached guess. Hover it and it explains why.

DROPAlso dropping the old sorting rule that pushed a model to the bottom of the table, and kept it out of the running for cheapest, whenever your output token count was higher than that model supports.