Voice AI cost is not a single line item. A voice agent has three cost layers: speech-to-text (STT), the LLM reasoning layer, and text-to-speech (TTS). Each has its own pricing. Add the telephony layer (inbound/outbound call handling) and you have four components to budget for.
The four cost layers of a voice AI call
Speech-to-text (STT): Deepgram is the dominant STT provider for real-time voice AI. Pricing runs $0.0059 per minute on the Nova-2 model — effectively negligible at most business call volumes. OpenAI Whisper is similar.
LLM layer: This is where cost varies most. GPT-4o Mini runs $0.15/$0.60 per million tokens (input/output). GPT-4o runs $2.50/$10.00 per million tokens. Claude 3.5 Haiku is $0.25/$1.25. A typical 3-minute call generates roughly 500–800 tokens of LLM context — meaning the LLM cost per call is $0.001–$0.008 for Haiku/4o-mini, or $0.01–$0.08 for GPT-4o.
Text-to-speech (TTS): ElevenLabs charges $0.30 per 1,000 characters on the Starter plan, dropping to $0.18 at scale. Cartesia is a strong alternative at a similar price point with lower latency. A 3-minute call with moderate speech output generates roughly 500–800 characters — putting TTS cost at $0.09–$0.24 per call on ElevenLabs.
Telephony: Twilio charges $0.0085 per minute for inbound calls plus $1.15/month per phone number. Telnyx is typically 30–40% cheaper. This is the most predictable layer — it scales linearly with call volume.
Platform pricing: VAPI, Retell AI, and Bland AI
These platforms bundle some of the component costs but charge a platform fee on top. VAPI charges approximately $0.05/minute as a platform fee on top of pass-through API costs. Retell AI is structured similarly at $0.07–$0.10/minute all-in for typical configurations. Bland AI runs $0.09/minute for standard calls.
All-in cost for a typical 3-minute call: $0.15–$0.35 depending on the LLM and TTS choices you make. At 200 calls per month, that is $30–$70 in running costs. At 2,000 calls per month, $300–$700.
How to estimate your monthly cost
Take your monthly inbound call volume, multiply by your average call duration in minutes, multiply by your blended per-minute rate (typically $0.08–$0.12 for a well-configured system). Add $50–$100/month for infrastructure (phone number, monitoring, logging).
Example: 500 calls per month, 3 minutes average, $0.10/minute blended rate = $150/month + $75 infrastructure = $225/month total running cost. Compare that to $3,500–$5,000/month for a human receptionist.
What is not included in these numbers
The build cost of the voice agent itself is separate — this is the engineering work to configure the call flow, write the intake scripts, build the CRM/calendar integrations, and test the system. For a custom build, that runs $10,000–$25,000. For a platform like VAPI with a DIY setup, you still need 20–60 hours of technical work.
Ongoing maintenance — script updates, integration fixes, model upgrades — typically runs 5–10 hours per month for a well-built system. Budget $500–$1,000/month if you outsource this, or internal engineering time if you have it.