which engines can it ask?
Two, perplexity sonar and claude, and every paid command defaults to claude. Each is asked with the key you bring and billed to you by the provider, and the library takes nothing on top.
why only two?
ChatGPT and Gemini are not wired in. Adding an engine is not the transport, it is a live key and a round against it, because a provider’s own account state decides whether the search tool is on, and a build that reports an engine it has never actually asked would be reporting on its own guess.
what does each model cost?
The published rate, as the provider’s own page stated it on the day it was read. A model with no price on file is refused rather than collected unpriced.
perplexity sonar
sonar ± $1 in and $1 out per million tokens, plus $5 to $12 per thousand requests, read 2026-07-31
sonar-pro ± $3 in and $15 out per million tokens, plus $6 to $14 per thousand requests, read 2026-07-31
sonar-reasoning-pro ± $2 in and $8 out per million tokens, plus $6 to $14 per thousand requests, read 2026-07-31
claude
claude-opus-5 ± $5 in and $25 out per million tokens, plus $10 per thousand searches, read 2026-08-01
claude-sonnet-5 ± $3 in and $15 out per million tokens, plus $10 per thousand searches, read 2026-08-01
claude-haiku-4-5 ± $1 in and $5 out per million tokens, plus $10 per thousand searches, read 2026-08-01