AI Voice Agent Cost Per Minute: Retell, Vapi, ElevenLabs

Every voice platform advertises a headline per-minute price, and none of them is the full cost. A call minute is made of four things: the platform's own fee, the language model, speech recognition and synthesis, and the phone line. Some platforms bundle some of those; others pass them through. This post adds them up from each vendor's own pricing page and shows the calculation.

I compared the three platforms on features and fit in Vapi vs Retell vs ElevenLabs. This is the cost companion to that post. Prices change often, so treat these as a worked method and check the pages before you commit.

What makes up the cost of a voice agent minute?

Every AI phone call minute has the same four layers, whichever platform you use.

• Platform fee. What the voice platform charges for orchestration: turn-taking, interruptions, call handling.
• Language model. The model that decides what to say. This varies the most, from fractions of a cent to several cents a minute; I compare the two biggest providers in Claude API vs OpenAI API.
• Speech. Speech-to-text to hear the caller and text-to-speech to reply. Some platforms bundle one or both into the platform fee.
• Telephony. The phone number and the per-minute carrier charge, from the platform's own numbers or a provider such as Twilio.

Retell: what does a minute cost?

Retell lists every layer on its pricing page, which makes the maths straightforward. Its voice infrastructure is $0.055 a minute. Telephony through Retell's own Twilio or Telnyx numbers is $0.015 a minute in the US, and numbers cost $2 a month. Text-to-speech starts at $0.015 a minute for Retell's platform voices and other standard providers, and ElevenLabs voices range from $0.040 (Flash) to $0.10 a minute. The model is priced per minute too.

Lean setup (GPT 4.1 mini at $0.0128, a standard voice):

• $0.055 + $0.0128 + $0.015 + $0.015 = $0.0978 per minute

Premium setup (Claude 4.6 Sonnet at $0.08, ElevenLabs Flash at $0.040):

• $0.055 + $0.08 + $0.040 + $0.015 = $0.19 per minute

Retell's own calculator example on the same page shows $0.11 a minute, which sits between the two. Optional add-ons add to this: a knowledge base is +$0.005 a minute, advanced denoising +$0.005, and PII removal +$0.01. Retell includes 20 concurrent calls free, with more at $8 per concurrent call per month.

Vapi: what does a minute cost?

Vapi charges a platform fee and passes provider costs through. Its pricing page lists "$0.05/min Vapi hosting" on the entry plan and says models pass through "at cost". The provider ranges it lists are Deepgram transcription at $0.0095 to $0.0099 a minute, OpenAI models at $0.0077 to $0.0452 a minute, and ElevenLabs voices at $0.0146 to $0.0238 a minute. Telephony is "charged by provider, not Vapi".

For telephony I use Twilio's US voice pricing: $0.0085 a minute to receive calls on a local number and $1.15 a month for the number.

Low end:

• $0.05 + $0.0095 + $0.0077 + $0.0146 + $0.0085 = $0.0903 per minute

High end:

• $0.05 + $0.0099 + $0.0452 + $0.0238 + $0.0085 = $0.1374 per minute

As a cross-check, Vapi's own calculator on that page estimates $82 to $129 for 1,000 minutes before phone numbers, which matches these figures without telephony. Watch concurrency on Vapi: the entry tier allows 4 concurrent calls, and extra lines are $10 each per month.

ElevenLabs Agents: what does a minute cost?

ElevenLabs bundles speech into one per-minute rate and bills the model separately. Its agents pricing page lists $0.08 per extra call minute on every plan, with plans that include minutes: for example, Pro is $99 a month with 1,238 minutes, which works out to $99 / 1,238 = $0.080 a minute. The page says the language model is billed separately "based on usage and varies by model", and that ElevenLabs adds no telephony fee: if you connect Twilio or a SIP trunk, that provider bills you directly.

• $0.08 + $0.0085 (Twilio inbound) = $0.0885 per minute, plus the model

Two details matter for real calls. The ElevenLabs cost help page says silence longer than 10 seconds gets a 95% discount, which helps on calls with long holds. And burst pricing, for calls above your concurrency limit, is charged at $0.16 a minute, double the standard rate.

What does a month of calls cost?

Per-minute prices are easier to judge as a monthly bill. Take 2,000 inbound minutes a month, for example 500 calls of four minutes each.

At this volume the platforms land close together. The bigger lever is the model and voice you choose: moving from a small model and a standard voice to a large model and a premium voice roughly doubles the Retell figure. Test whether the cheaper option sounds and behaves well enough for your callers before paying for the premium one.

What these numbers leave out

The per-minute figure is the running cost of the call itself. A realistic budget also covers:

• Build and testing. Prompt design, integrations with your calendar or CRM, and test calls, priced like any AI developer engagement. Testing is a project in itself; I will cover it in a later post on go-live testing.
• Integrations. If the agent books appointments or looks up orders, those systems and any workflow tool have their own costs. My comparison of n8n, Zapier and Make covers that side.
• Compliance. Vapi, for example, lists HIPAA support as a $2,000-a-month add-on on its pricing page.
• Human handoff. Transferred calls keep using telephony minutes while a person takes over.

For when a voice agent is worth it at all, and how to design one for support, see my guide to AI voice agents for customer service.

Work with me

I build voice agents on Retell, Vapi and ElevenLabs, and pick the stack by call volume, latency and cost per minute rather than habit. See voice AI agents and AI agent development, or browse my work and book a scoping call.