What a minute of your voice agent costs — and which layer is running the bill.
Three meters run at once on every call — speech-to-text, the model, text-to-speech — plus telephony and a platform fee, all per second. Connect an orchestrator (Vapi, Retell) for per-call breakdowns, the component vendors for their meters, or price a stack below before you build it.
Calculator — price a stack per minute
$0.123 a minute · $614 a month
STT $0.008 · model $0.006 · TTS $0.045 (scaled by talk share) · telephony $0.014 · platform $0.050. Rates are public list prices, illustrative; the invoice wins.
How voice bills run away
The agent talks too much. TTS is per character; a chatty system prompt doubles the most expensive layer. The prompt grows every turn. Tokens per minute climb through a call; a running digest caps it. Silence is billed. STT charges for hold music and dead air; transport charges both legs. Short calls cost real money. Voicemail and hang-ups still pay for connection, greeting, and analysis. Analysis on every call. Summaries and structured extraction are a second model call per conversation.