Latency and cost

Get replies faster. Understand what takes time and what each minute costs.

The assistant editor estimates the reply time and cost per minute for your exact settings, split into listen, think and speak. After each call, the call page shows the measured time of every turn.

#Where the time goes

Part	Typical	How to reduce it
End of turn	200 to 500 ms	Lower `eot_threshold` and keep `eager` on. See Speech-to-text.
First token from the model	250 to 400 ms	Keep `priority` and `race` on. Shorter instructions help a little.
First sound from the voice	100 to 500 ms	Cartesia and ElevenLabs Flash start fastest. See Voices.
Tools	depends on your API	Set a start message so the caller hears something while a slow tool runs.

#Cost

Pay the monthly platform fee plus your actual AI calling usage, at $0.14/min on Core or $0.08/min on Pro. No calling minutes are included in the platform fee. Browser test calls use a separate test allowance. See Plans and limits.