Which voice agents are the cheapest?
Which voice agents are the cheapest? There is no verified universal cheapest option because providers include different things in their prices. Vapi advertises a $0.05-per-minute hosting charge with model costs passed through separately. Retell AI advertises a $0.07–$0.31-per-minute AI voice range. Neither number alone establishes the lowest complete bill. Compare the same call workload, configuration and operating requirements, then validate the estimate with actual usage.
Read the unit before comparing the number
A hosting minute, AI minute, included credit and completed call are different billing units. Check which charges cover orchestration, speech, language models and telephone service. Also check whether a subscription or minimum commitment is required to obtain the advertised rate.
ElevenLabs documents separate LLM charges for its agents and explains that voice-call billing follows connection duration. That illustrates why a displayed agent allowance is not necessarily your complete running cost.
For VoiceAIWrapper’s native engine, the published planning range for additional AI calling is approximately $0.07–$0.31 per minute, depending on models and voice. Telephony is extra. BYOK usage is billed by the connected provider; do not also count native usage for those same external-provider calls. Platform or BYOK add-on charges can still apply.
Compare three identical workload scenarios
Use a small, medium and agency workload so a low introductory price does not dominate the decision. These are illustrative calculations, not vendor quotes:
- Light use: 100 calls averaging three minutes equals 300 minutes. Vapi’s advertised hosting component would be $15, before models and other charges. Retell’s advertised AI range would imply $21–$93, before separately applicable charges. These figures cover different components and cannot be ranked as complete totals.
- Growing use: 1,000 calls averaging three minutes equals 3,000 minutes. The same calculations produce $150 for Vapi hosting and $210–$930 using Retell’s AI range. Obtain configuration-specific estimates instead of assuming the bottom of a range applies to your agent.
- Agency use: ten clients each handling 100 three-minute calls also total 3,000 minutes. The conversation volume is unchanged, but client access, separate reporting, billing and support create additional requirements. Compare the cost of meeting those requirements on every shortlisted platform.
Use the same telephony route, average duration, language and required actions across quotes. If a provider does not disclose a component, mark it unknown rather than zero.
Calculate the monthly bill without double-counting
Start with the subscription or minimum charge. Add usage beyond included allowances, telephony, phone-number rental, required add-ons and any separately charged support. Keep taxes and one-time implementation work visible.
Apply included credits once. Do not charge every minute again at an overage rate when part of the workload is already covered. Equally, do not divide a bundled agency subscription by an approximate included minute allowance and call the result a pure AI rate: that subscription also buys platform capabilities.
Check credit expiry, rollover, billing increments and the treatment of transfers, unanswered calls and testing. These details can change the total even when two headline prices look similar. The VoiceAIWrapper cost guide explains its platform, AI and telephony components in more detail.
VAW’s model picker shows a per-minute model cost before selection, and each call provides line items for model, voice, transcription, platform and knowledge-base costs. Reconcile those with telephony billed through your own Twilio account rather than treating the picker’s number as the entire bill. When VAW credits run out, new calls stop while calls already in progress finish; check auto top-up settings before budgeting. Closing phrases, a silence timeout and a maximum call length can also help limit idle call time when configured appropriately.
Include the cost of getting the job done
The cheapest minute may not produce the cheapest completed task. Track how many calls produce the correct booking, answer or handoff, and how much staff time is needed to repair mistakes. Include setup and maintenance effort consistently across candidates.
For example, dividing the same total running cost by fewer successfully completed tasks produces a higher cost per useful outcome. This is arithmetic, not evidence that one provider performs better. Measure completion with your own representative test calls before using it to justify a purchase.
Avoid cutting model capability solely to reduce the rate. Try a less expensive configuration against the same acceptance tests and keep it only if it still meets your requirements.
Choose the lowest verified total for your workload
Request an itemized estimate, run a controlled pilot and reconcile the actual bill against your assumptions. Compare an ordinary month after welcome credits as well as the first month. For annual commitments, consider the cash commitment and unused allowance rather than only the monthly equivalent.
The cheapest suitable option is the one with the lowest verified total while meeting your quality, integration and support requirements. Until those conditions match, a numerical ranking would create false precision.
Pricing references checked 18 October 2026; confirm current terms before purchasing.