speech-to-speech calculator

Realtime voice costs both ways.

Model native audio-in/audio-out systems separately from an STT → LLM → TTS pipeline. Adjust the caller-to-agent talk ratio and see every material layer.

Realtime speech modelPlatform / orchestrationCarrier / transportRecording

Estimated monthly total$0

$0/min · $0/call

platform$0
caller audio input$0
agent audio output$0
carrier$0
recording$0

Model assumption

official model pricing ↗

how to read it

Connected time is not billed audio time

A five-minute call contains caller speech, generated agent speech, and silence. Realtime providers may bill input and output audio at different token rates, while the platform and carrier usually bill connected time.

model cost = caller audio minutes × input rate + agent audio minutes × output rate

monthly total = platform + realtime model + carrier + recording

OpenAI entries convert published audio-token prices into audio-minute estimates using 600 input audio tokens and 1,200 output audio tokens per minute. Actual bills also depend on retained conversation context, cached input, text and reasoning tokens, interruptions, silence handling, and provider-side truncation. Replace these planning assumptions with usage-log measurements before committing spend.

primary sources

Current realtime model pricing

Public model pricing checked 2026-07-18. Preview model rates and names can change without notice.