Reading your Usage & Savings page

The Usage & Savings page rolls your metered requests up over a date range you choose. The hero cards summarize the range; All metrics below breaks it down in full. The page has two lenses: Cost $ for API-key accounts and Quota % for subscription plans. It opens on the one that matches your traffic and remembers your choice.

The hero cards (Cost $ lens)#

• Cost saved: total saved over the range, and what share of spend that represents.
• Spend: what you actually paid, after Tidbit, with the prior period shown as "was $X".
• Cache hit rate: the share of your tokens served from cache rather than processed fresh. Cache writes count as fresh, because that is how they are billed.
• Tokens cut: redundant payload removed before the request went out, estimated from what was dropped.

A range with a lot of cache writes in it will show a low hit rate and a high cost, and that is the honest reading rather than a fault. The first request against a new prefix pays full price to write the cache; the requests after it read that prefix back cheaply. Cache write tokens are broken out under All metrics so you can see when a range is carrying that cost.

Ratios need a little traffic before they mean anything, so Cache hit rate and Work unlocked read "Not enough data yet" until the range holds at least five requests and a thousand tokens. A multiplier computed from one request is noise, and showing it to one decimal place would dress it up as a measurement.

Trend deltas compare the selected range against the previous equal-length window, so a 30-day view is measured against the 30 days before it. Today runs from midnight in your own timezone; the longer ranges are counted in UTC and say so next to the range.