Voice AI agent cost calculator
Self-hosting removes the per-minute platform fee, but you pay for your own server. Put in your numbers to see where that crosses over. Prices were read from each provider's own page on 6 November 2026, and they do change.
$0.005/min audio in plus $0.018/min audio out, as if audio ran both ways for the whole call. Treat it as a high estimate.
Your self-hosted month
Platform fees for the same minutes
These cover the platform fee only. You'd still pay for models and phone minutes on top.
Prices used
| Phone provider | Inbound | Outbound | Streaming | Number |
|---|---|---|---|---|
| Twilio | $0.0085/min | $0.014/min | $0.0044/min | $1.15/mo |
| Telnyx | $0.0052/min | $0.007/min | $0.0035/min | $1.00/mo |
- Twilio: Local US number.
- Telnyx: Voice API plus starting SIP trunking rates, local US number. SMS adds $0.10 a month per number.
- Google Gemini Live: $0.005/min audio in plus $0.018/min audio out, as if audio ran both ways for the whole call. Treat it as a high estimate.
- xAI Grok voice: grok-voice-3.0 speech to speech.
- OpenAI Realtime (enter your own): OpenAI prices its realtime models per token, with no per-minute figure. For example, gpt-realtime-2.1-mini is $10 per 1M audio tokens in and $20 out. Check your usage to estimate a rate.
- Vapi hosting: $0.05/min.
- Retell voice infrastructure: $0.055/min.
What this leaves out
Your own time, taxes, SMS, and LiveKit Cloud if you use it instead of running the open-source LiveKit server. The server figure is yours to fill in, because it depends on where you host and how many calls run at once.
Ready to try it? Start with the Quick Start, or read open-source alternatives to Vapi and Retell.