MundoKit

Voice Agent Cost Calculator

Choose session-time or audio-token billing and enter billed audio shares, billed text and backend costs.

Inputs
Used only for VAD-filtered input; overlaps are allowed.
Used only for manual input. Include committed silence.
Example share.
gpt-realtime-2.1 preset as of 2026-11-14.
gpt-realtime-2.1 preset.
Conversation replay beyond newly spoken audio. Uncached estimate.
Include repeated instructions and history; example.
Example.
gpt-realtime-2.1 preset.
gpt-realtime-2.1 preset.
Enter separate backend, telephony, transcription and tool charges.

Results

Total estimated cost (USD)
$40.00
Per session
$0.40
Per session minute
$0.04
Per-session breakdown
ComponentUSD / tokens
Voice charge$0.38
Text charge$0.02
Backend/tools$0.00
Input audio tokens2,400
Output audio tokens4,800
  • Input assumes VAD-filtered speech; manually committed silence is excluded from this scenario.
  • Prices checked 2026-11-14; all rates are editable. Audio conversions are estimates. Replayed history, cache discounts, special tokens and actual speech detection change bills. Enter measured billed totals where possible. Session mode charges elapsed time regardless of speech share.

Results update as you type. The share link stores your inputs after the # in the address, which is not sent to our server, but anyone you send it to can read them.

Prices change often. We list the date we last checked each price; confirm on the provider’s official pricing page before you budget.

How it works

Editable presets checked 2026-11-14: GPT-Live-1 costs $0.05 per session minute, billed per second. Backend usage is separate. Realtime presets use gpt-realtime-2.1: audio $32 input/$64 output and text $4 input/$24 output per million tokens.

audio tokens = minutes × 60 × billed audio share × tokens/second
session total = voice charge + text charge + backend charge

OpenAI documents about 10 input and 20 output audio tokens per second. By default, input uses VAD-filtered speaking share. For manually committed input, choose manual mode and enter the share of session audio committed, including silence. Output uses the agent speaking share. Replayed conversation input can add charges; enter it separately. Rates here assume uncached input. Multiply per-session cost by session count. Do not round intermediate values; displayed USD uses cents or significant digits for small amounts.

Worked examples

Token-priced ten-minute session

With VAD-filtered input and 40% speaking for each side, audio uses 2,400 input and 4,800 output tokens. Audio costs $0.384; 1,000 text input and 500 text output tokens add $0.016. Total is $0.40 per session or $40 for 100 sessions.

Session-time billing

Ten minutes × $0.05 = $0.50 per session. For 100 sessions, voice costs $50; a separate $0.10 backend charge per session brings the total to $60.

Frequently asked questions

Does silence cost money?
Session-time mode bills elapsed session time. VAD filters empty input audio unless the client manually adds it as conversation input. Manual mode counts committed audio, including silence; VAD mode counts user speech. Replay and special tokens can add charges.
Why can actual realtime cost be higher?
History is sent again on later turns. Include billed replay and repeated text, and inspect usage events.
Are cache discounts included?
No. These estimates use uncached input rates; effective rates can be entered manually.
Does backend cost include telephony?
Only if you enter it. Add telephony, tools and other backend charges to the per-session amount.

Sources

Last reviewed 2026-11-14; next review due by 2026-12-14. The method is written out above so you can check our math.