Voice Agent Cost: Adding Up STT, the LLM and TTS
A voice agent is really three services in a loop: transcribe the caller, reason with an LLM, and speak the reply. Its cost is the sum of all three, so estimating one piece badly understates the whole.
Three meters running at once
Every minute of conversation bills speech-to-text on the audio, the LLM on the tokens exchanged, and text-to-speech on the spoken reply. Add them per minute to get a realistic cost, rather than pricing only the model.
Where to trim
Keep replies concise to cut both LLM output and TTS characters, and pick efficient STT and voice tiers. Because three meters compound, small savings on each add up quickly across a long call.
Calculators for this topic
Frequently asked questions
What does a voice agent cost per minute?+
The sum of transcription, LLM and speech-synthesis cost for that minute — estimate all three, not just the model.
How do I make a voice agent cheaper?+
Keep spoken replies short and choose efficient STT and TTS tiers, since all three costs run every minute.