AI voice agent cost: per-minute pricing, build costs and a worked example

An AI voice agent built on a voice platform typically costs somewhere around $0.09 to $0.14 per minute in usage fees, from public price pages we checked on 2026-10-15. That figure adds telephony, speech-to-text, a language model, text-to-speech and the platform's own fee. It excludes the one-time build, ongoing monitoring and the cost of the calls the agent hands to a person. Scenox does not publish prices, so below we break the stack down, show how to calculate cost per call, and walk through a worked example with every assumption stated.

#The five components of cost per minute

A call to an AI agent passes through five services. Each one bills separately, or the platform bundles them.

1. Telephony: the phone number and the minutes carried over the phone network.
2. Speech-to-text (STT), which turns the caller's speech into text as they talk.
3. The language model (LLM), which decides what to say and which tool to call, such as checking a calendar.
4. Text-to-speech (TTS), which turns the reply into a voice.
5. The orchestration platform, which handles turn-taking, interruptions, call transfer and logging.

You can buy these as one product or assemble them yourself. Bundled products start faster and show you less. An assembled stack gives you control over each part and leaves more for you to maintain.

#What the public price pages say

We read these pages on 2026-10-15. Prices change often and vary by plan, voice, model and country, so treat the table as a snapshot.

Adding the Vapi-listed components gives a low case of about $0.09 per minute (cheapest listed STT, LLM and TTS, inbound telephony) and a high case of about $0.14 (most expensive listed, outbound). Retell's range is wider because it includes premium models and voices. Vapi's own estimator shows $82 to $129 for 1,000 minutes a month on its example configuration, which includes a $50 hosting fee.

Two cautions. Add-ons such as call recording analysis, noise removal and quality scoring can add per-minute fees. And check how each platform counts a billable minute, including whether ringing, hold or transfer time counts.

#From cost per minute to cost per call

Cost per call is minutes times rate, so call length matters more than any single component.

At $0.09 to $0.14 a minute, a 4-minute call costs about $0.36 to $0.57 in usage. A 12-minute call costs three times that. Long calls come from agents that ask too many questions, repeat themselves or take too long to confirm. Tight call design cuts cost and gives callers a better experience.

#Build and run costs beyond usage

Usage fees are the visible part. Four other lines decide whether the project pays back.

• Build: call-flow design, a knowledge source, calendar or CRM integration, escalation rules, voice and tone testing, and testing against recorded calls. A simple inbound receptionist is less work than an outbound agent that books across several systems.
• Monitoring. Someone listens to a sample of real calls each week, reads transcripts of failures, and updates the flow and knowledge. Most proposals skip this line.
• Human fallback. Calls the agent transfers or can't finish still cost staff time.
• Compliance. Recording consent, disclosure that the caller is talking to an AI, and data retention rules differ by country and state. Get legal advice for your markets. We don't give it here.

#A worked example (illustrative)

A property management office receives 1,500 inbound calls a month about maintenance requests, viewings and rent queries. Average call: 4 minutes. A person handles each call in 6 minutes including notes, at an assumed $25 an hour loaded cost, which is $2.50 a call. These are assumptions, not benchmarks.

Assume the agent fully resolves 70% of calls and transfers 30% to a person.

The result is mixed. At 1,500 calls a month, with heavy monitoring and premium components, the agent costs about what people cost, and slightly more. With lean monitoring it saves a few hundred to well over a thousand dollars a month. That is before any build cost, which for a project like this might run from 60 to 200 hours (about $6,000 to $40,000 at an assumed $100 to $200 an hour).

The case improves with more calls, shorter calls and a higher resolution rate. It also improves with value the table doesn't capture: calls answered at 2 a.m., fewer missed viewings, faster repair dispatch. Low volume kills it. If your office takes 200 calls a month, a voice agent is rarely worth building.

Key point: The per-minute price is the smallest decision in a voice agent project. Call length, resolution rate and monitoring effort decide whether it pays back, so ask a vendor to model those, not only the usage fee.

#Inbound and outbound calls cost differently

Inbound agents answer calls people chose to make, so callers are motivated and calls run shorter. Outbound agents, such as appointment reminders or lead follow-up, cost more per minute on the telephony line ($0.014 against $0.008 to $0.0085 inbound on the US pages we read). They also carry more compliance risk. Consent rules for automated and AI-voiced calls vary by country and state, and a mistake there costs far more than a few cents a minute. We wouldn't start with outbound cold calling.

#Quality choices that move the price

Cheaper components can lower the bill and raise the number of failed calls. A slower model makes the agent pause before it answers, and callers read that as a bad line. A cheaper voice can sound flat by minute five. A narrow speech model can mishear names and addresses, which hurts when the agent is booking a repair visit. Test each choice on recorded calls from your own line, and track resolution rate next to cost per minute.

#How to keep voice agent cost down

1. Shorten the call. Confirm the caller's need in the first 20 seconds and skip the scripted preamble.
2. Match the model to the task. Booking and FAQ calls rarely need the most expensive model, so test a smaller one on recorded calls.
3. Pick the voice for the use case. Premium voices cost more per minute. Check whether callers notice.
4. Cap runaway calls with a maximum duration and a transfer rule.
5. Send anything risky to a person. Payments, complaints and legal questions go to staff.
6. Review a weekly sample of calls. It surfaces cost and quality problems together.

#AI receptionist or custom voice agent

An AI receptionist product is a packaged voice agent for answering, qualifying and booking. It suits standard needs and small teams. A custom voice agent earns its cost when calls touch your own systems, such as a maintenance ticketing system or an ERP, or when you want control over scripts, escalation and where recordings live. Our voice agents page explains what we build. For build versus run, see AI automation cost, and use the ROI method to test a specific case. Our sample audit shows how a workflow is sized before anything gets built.

If you want call volumes, durations and systems turned into a realistic cost model, book a strategy call.

#Frequently asked questions

#How much does an AI voice agent cost per minute?

On the public price pages we checked, a platform-built agent comes to roughly $0.09 to $0.14 per minute when you add the platform fee, speech-to-text, a language model, text-to-speech and telephony. One all-in platform lists $0.07 to $0.31 per minute depending on model and voice. Verify current prices before budgeting.

#How much does an AI receptionist cost?

Usage fees for a 4-minute call are around $0.36 to $0.57 on the stack above. The full cost includes setup, monitoring and the calls a person still handles. Packaged AI receptionist products price differently, often per plan or per call, so compare the plan page against your call volume.

#What is the cheapest way to run a voice agent?

Shorter calls, a smaller model for simple tasks, a standard voice, and lean monitoring. Self-assembling the stack can reduce unit costs, and it raises the engineering you must maintain.

#Does a voice agent replace a receptionist?

It can answer routine calls, book appointments and take messages. It should transfer complex, sensitive or upset-caller calls to a person. Most teams use it to cover after-hours calls and peaks, not to remove the front desk.

#Is cost per call or cost per minute the better measure?

Cost per resolved call. It includes transfers, retries and monitoring, so it shows what a completed customer request costs.