An AI voice agent costs about $0.10 to $0.15 a minute to actually run. That's four commodity services, telephony, speech-to-text, a language model, and text-to-speech, plus a thin layer to orchestrate them. Managed platforms charge $0.25 to $1.00 a minute for the same loop. The gap, two to ten times the real cost, is middleman markup on your own phone calls.
Here's the arithmetic every "AI voice agent pricing" guide avoids, because the companies writing them are the ones collecting the markup.
How an AI voice agent works (and what each piece costs)
An AI voice agent answers a live call end to end. The loop runs sub-second, over and over, for the whole conversation:
- Telephony carries the call audio (a phone number and carrier).
- Speech-to-text transcribes the caller in real time.
- A language model figures out intent and decides the reply, and can hit your CRM or calendar.
- Text-to-speech speaks the answer back in a natural voice.
Four metered commodities and a bit of glue. Now the part the cost guides bury: each of those four has a public, published price.
What an AI voice minute actually costs
Assemble it yourself and here's the real floor, from each provider's own pricing:
| Layer | At-cost price |
|---|---|
| Orchestration (Vapi) | ~$0.05/min (STT/LLM/TTS passed through at cost) |
| Speech-to-text (Deepgram) | ~$0.005/min |
| Language model (fast/mini) | ~$0.003 to $0.03/min |
| Text-to-speech | ~$0.02 to $0.03/min |
| Telephony (Twilio) | ~$0.014/min + $1.15/mo per number |
| All in | ~$0.10 to $0.15/min |
Vapi is the tell: it prices the orchestration layer at a nickel and explicitly passes the rest through at cost. The four commodities under the hood total about a dime a minute.
What the platforms charge you
Now the rate cards. From vendors' own pricing pages:
- Retell: roughly $0.07 to $0.31 a minute
- Aircall: about $0.49 a minute
- CloudTalk: about $0.50 a minute pay-as-you-go
- JustCall: about $0.99 a minute
The honest one here is Retell, which itemizes its own stack, voice infra, TTS, LLM, telephony, adding up to roughly nine cents of real cost sitting under a rate card that climbs past thirty. Everyone billing $0.25 to $1.00 a minute is charging two to ten times the underlying cost. At 5,000 minutes a month, that's the difference between a real cost near $500 and an invoice of $1,250 to $2,500. The delta is pure middleman rent, a growth tax on your own phone calls.
Are AI voice agents worth it?
Yes, and that's exactly why the markup matters. The competitor guides all run the same ROI math: a human answering service runs $3,000 to $4,000 a month all in, an AI voice agent runs a fraction of that, and it pays for itself in a month or two. That's real.
The twist they leave out: the AI is even cheaper than they admit, once you strip the rebill. The agent that answers every call, qualifies the lead, and books the appointment is a great deal at a dime a minute. It's a worse deal at fifty cents, and the only thing the extra forty cents buys is the platform's dashboard and their right to raise your rate.
The point: own the wiring
Every ranking page warns you about "hidden fees," meaning a competitor's fees, while never disclosing that the four services under the hood total about ten cents a minute. None of them tell you the alternative: connect the telephony, speech, model, and voice layers directly and pay cost. It's the same thesis we ran on chat in AI chatbot development services, the model is the cheap part, the markup is the product.
The Allodra model wires the voice agent straight to the providers inside a CRM you own outright. The agent answers inbound, qualifies the caller, and books into the same pipeline your automations run on, and the only per-minute cost is what the commodities actually charge. No platform sitting in the middle of every call with a meter running.
The bottom line
AI voice agent pricing is quoted per minute, at two to ten times what the minute actually costs, because the companies quoting it resell the minute. The underlying AI is a commodity at about a dime. The markup is the business model.
Before you sign a per-minute contract, run your call volume through the cost analyzer and compare the rented rate against owning the wiring. And if you're an agency that wants to capture that markup instead of paying it, that's a different play, we covered it in the white-label voice agent breakdown.