How to build a complete voice agent call budget

IC Guides · AI Agents · Build an agent

By IC Lab Desk · Edited by Yajush Gupta · How we report

Use this guide to build a complete voice-call budget, from hosting to telephony. For a platform decision using one matched workload, read our Vapi vs Retell cost comparison.

Vapi and Retell AI both sell the plumbing for AI phone agents and both charge by the minute, but they split the bill differently. Vapi charges $0.05 a minute for hosting and passes the transcriber, model, voice and (usually) the phone carrier through at cost. Retell publishes one stack of components, each with its per minute price, and its example comes to $0.11 a minute. Neither headline is the cost of a call. Add the layers yourself.

Who it fits

Developer teams building AI receptionists, outbound calling or phone support. Both are platforms you configure and test, not finished products. If you want a vendor to run the whole support workflow, read quote only support agents instead.

What Vapi publishes

Item	Price
Hosting, no success package	$0.05 per minute, prepaid. $5 free credits to start. 4 concurrent calls, 2 organizations, 1 phone number, 14 days of raw data retention.
Core package	$29 a month. 10 concurrent calls, 5 phone numbers, 30 days retention, zero data retention option, 2 business day email response.
Pro package	10% of the Vapi hosting fee, with a $999 a month minimum. 30 concurrent calls, 10 organizations, 180 days retention, role based access, dedicated Slack support, 99% uptime SLA.
Premier	Custom, sales only. 99.9% uptime SLA, SSO, dedicated SIP.
Add ons	HIPAA handling $2,000 a month. Extra concurrent line $10 a month. Extra organization $20 a month.
Transport	Vapi telephony and SIP, WebSockets and Daily WebRTC are free. Twilio $0.008 inbound and $0.014 outbound a minute, Vonage $0.00814, Telnyx $0.0055, charged by the provider.
Models, voice, transcription	Provider cost with no markup, per Vapi. Its calculator shows Deepgram at about $0.0095 to $0.0099 a minute, OpenAI at $0.0077 to $0.0452 and ElevenLabs at $0.0146 to $0.0238.

What Retell publishes

Item	Price
Headline	$0.07 to $0.31 a minute for voice agents, from $0.002 a message for chat agents. $10 in free credits. No platform fee.
Voice infrastructure	$0.055 a minute.
Text to speech	$0.015 a minute on Retell, Minimax, Fish, Cartesia, OpenAI and Inworld voices. ElevenLabs runs $0.04 to $0.10 depending on the model.
Language model	Billed per minute by model, from $0.0016 (GPT 5 nano) up to $0.64 (GPT 6 Astra, fast tier). The models Retell marks as recommended sit at $0.064 to $0.16 on the standard tier.
Telephony	$0.015 a minute on Retell Twilio or Telnyx numbers. No charge for your SIP trunk.
Add ons, per minute	Knowledge base $0.005. Advanced denoising $0.005. Safety guardrails $0.005. PII removal $0.01. AI quality assurance $0.10 after 100 free minutes.
Monthly items	Phone number $2. Extra concurrency $8 a line beyond the first 20 free. Verified phone number $10. Knowledge bases $8 each after the first 10.
Other per event fees	Batch calling $0.005 a dial. Branded caller ID $0.10 per outbound call. SMS $0.01 a message.

Worked examples (our arithmetic)

Retell’s calculator default adds up to $0.11 a minute, made up of $0.04 for the model, $0.055 for infrastructure and $0.015 for voice, with telephony at zero. If you buy the number from Retell and use Twilio, add $0.015 and the figure is $0.125 a minute, or $125 for 1,000 minutes.

Vapi’s calculator for the same 1,000 minutes shows hosting $50, Deepgram $10, OpenAI $8 to $45 and ElevenLabs $15 to $24, which is $82 to $129 in total before rounding, or $0.082 to $0.129 a minute. Put a Twilio outbound call on top at $0.014 and the range moves to roughly $96 to $143 for 1,000 minutes. The two platforms land in the same band once you stack the components. The choice of model and voice creates a larger cost gap than the choice of platform.

Fixed fees change the picture at the edges. Vapi Pro is 10% of the hosting fee with a $999 floor, so the floor only stops binding when hosting reaches $9,990 a month, which is 199,800 minutes at $0.05. Below that you pay the $999. On Retell there is no equivalent package floor, and concurrency is cheaper, 30 lines above the free 20 costs $240 a month at $8 each.

What is not confirmed

We could not read from Retell’s page which compliance items (HIPAA, BAA, SSO) apply to the pay as you go plan and which need Enterprise, because the comparison table marks them with symbols that did not come through in our extraction. Ask before you assume.

Vapi’s FAQ answers were collapsed on the page, so what exactly counts inside the per minute cost and how the Pro fee is calculated come from the headline text only. Neither vendor publishes call quality, latency under load or failure rates on the pricing page, and we did not test either.

Where costs surprise you

Silence is billed. Retell states billing covers the entire call, including silence and hold, because speech to text stays active. It bills to the nearest second, and failed calls to connect are not charged. After a transfer the AI fee stops but telephony continues.

Model choice moves the bill more than platform choice. On Retell the same call costs $0.0016 a minute on the smallest model and $0.64 on the largest fast tier model. Add ons are priced per minute too, and knowledge base, guardrails, denoising and PII removal together add $0.025 a minute on Retell, which is about 23% on top of the $0.11 example.

Compliance is a step change on Vapi, where HIPAA handling is $2,000 a month before you have taken a call. Vapi’s hosting is also prepaid, so cash is tied up ahead of usage.

How to test it

1. Pick one call script, one voice and one model, and build it on both platforms.
2. Place 50 test calls of realistic length, including some with silence and a transfer.
3. Read the invoice line by line and divide by connected minutes. That is your real cost per minute.
4. Repeat at your expected peak concurrency to see whether you hit the free line limits (4 on Vapi without a package, 20 on Retell).

Questions to ask before you buy

1. What is the total per minute cost for our exact stack, including carrier fees?
2. Which compliance features sit on which plan, and at what price?
3. What happens to concurrency at our peak hour, and what do extra lines cost?
4. How are transfers, voicemail and very short calls billed?
5. What is the notice period if a model provider raises prices?

Sources

• Retell AI, Pricing, retellai.com/pricing, checked 12 October 2026.
• Vapi, Pricing, vapi.ai/pricing, checked 12 October 2026.

Worked examples are IC arithmetic from the published rates. We did not test the products. Prices change, so check the date. See the live rows in the AI Agent Pricing Tracker. Send a correction.

Related guides Vapi vs Retell · Workflow automation pricing · OpenAI Agents API cost · AI agent glossary · Which AI agent, and what it costs · AI Agent Pricing Tracker · All AI agent guides