What does a minute of voice AI actually cost?
A neutral, sourced index of speech-to-text, LLM, text-to-speech, realtime and telephony prices, from 21 vendor pricing pages. Build a stack, see the real per-minute cost, and compare platforms on the same basis.
Cost-per-minute calculator
Assumptions (edit to match your calls)
* Cheapest verified realtime cascade at default assumptions: Soniox stt-rt-v5 (real-time) ($0.0020) + OpenAI gpt-5-nano ($0.0005) + Google Gemini API gemini-3.8-flash-lite-tts ($0.0045) + Telnyx Elastic SIP, local inbound ($0.0032). Excludes orchestration/hosting compute (e.g. self-hosted Pipecat or LiveKit Agents servers), phone number rental and taxes. Promotional prices included where flagged.
Browse by layer
Speech-to-text
Streaming transcription, billed on the whole call.
17 items →LLM
Per-token prices converted to a per-call-minute estimate using the assumptions below.
22 items →Text-to-speech
Billed on generated speech only, so scaled by agent talk share.
21 items →Speech-to-speech / realtime
Single models that hear and speak directly.
5 items →All-in platforms & orchestrators
Hosting or bundled per-minute fees. Read what each includes.
11 items →Telephony (US)
Carrier minutes. Inbound and outbound differ.
13 items →Popular comparisons
Vapi vs Retell pricing
Same models, same assumptions: what a minute really costs on Vapi and Retell.
Deepgram vs AssemblyAI pricing
Streaming speech-to-text and voice agent API prices side by side.
ElevenLabs vs Cartesia pricing
Per-character vs credit plans, converted to cost per call-minute.
Twilio vs Telnyx vs Plivo
US inbound/outbound per-minute and number rental compared.