what does a sonar call cost?
sonar is the model a command asks on perplexity when no --model is given, and the one lulu doctor spends its single test call on. Its price is a band rather than a number, because most of the bill is a request fee that depends on the search context size.
what is the published rate?
sonar ± $1 in and $1 out per million tokens, plus $5 to $12 per thousand requests.
Read 2026-07-31 from docs.perplexity.ai/getting-started/pricing. It is read off the table the library bills against, and a test fails if the two differ.
where does the money go?
A short sonar call costs a fraction of a cent in tokens and between half a cent and just over one cent in fees. For a search-grounded model the fee is not a rounding error next to the tokens, so the floor, the fees with no token counted, is most of the bill. Collapsing the band to one number would mean choosing a search context size on the user's behalf and then presenting the guess as an invoice.
That is why sample design is a cost decision and not a statistical nicety. A measurement of 40 prompts asked 5 times each is 200 calls, which is $1.00 to $2.40 in request fees plus a few cents of tokens.
why is it never priced like the next model up?
Because the nearest relative is where a fifteen-fold error comes from: sonar and sonar-pro differ by that much on output tokens. A model whose price has not been checked gets "no published price on file" rather than the price of its nearest relative, and every recorded call is priced at the rate of the model that answered it, read off the record rather than off a flag.
what else does this engine price?
how does it ask perplexity sonar? ± every engine and every price