What is the cheapest voice AI platform?

On published headline rate, Vapi at 5¢ per minute, then Bolna at 6¢. But Vapi’s figure excludes speech and model costs, which are passed through, so the cheapest headline is not reliably the cheapest bill. Model your own volume and model choices across vendors before deciding.

Are phone numbers included in these prices?

It varies, and it is often the component people forget. Rexa.ai includes phone numbers in its per-minute rate by default, provisioning the number itself, and also supports bringing your own carrier account — on that path your carrier bills you directly. Bolna includes a free number for the first 30 days on its Pilots plan. Deepgram does not provide telephony at all. Vapi’s pricing page prices concurrency per line but does not state number provisioning.

Is a voice AI agent cheaper than a human caller?

Per minute, substantially — but that comparison flatters the agent, because a person handles complexity an agent escalates. The honest comparison is cost per resolved outcome across the whole queue, counting the calls that still reach a human. For high-volume repetitive calling the gap is large; for complex or high-value conversations it narrows quickly.

Learn

How much does a voice AI agent cost?

The cost of a voice AI agent is the sum of telephony, speech recognition, language model, speech synthesis and platform fees, billed per minute of conversation.

The short answer

Published rates in 2026 run from roughly 5¢ to 16¢ per minute, but the headline number rarely covers the same things twice. Some platforms quote a platform fee and pass speech and model costs through separately; others bundle everything including the phone line. To compare honestly, work out your total cost per connected minute — platform fee plus every pass-through component plus telephony — rather than comparing headline rates.

Last reviewed August 8, 2026

What are the cost components?

Five. Telephony (the call itself and the phone number), speech-to-text, the language model, text-to-speech, and the platform fee for orchestration, recording, dashboards and compliance. Every vendor charges for all five; they differ in how many are folded into one quoted number.

Why do headline rates differ so much?

Because they measure different scopes. As of the access dates below: Vapi publishes 5¢ per minute as a platform fee with model costs passed through at cost. Bolna publishes 6¢ per minute with volume discounts and supports bring-your-own-keys. Deepgram’s Voice Agent API is 7.5¢ per minute standard and 16.3¢ advanced, and does not include telephony at all. Rexa.ai publishes 10¢ per minute with speech, model, voice, recordings and — on its default setup, where it provisions the number — the phone line included; bringing your own carrier account is supported instead, in which case that carrier bills you separately. Synthflow does not publish per-minute pricing; its enterprise contracts start at 30,000 annually.

How should you compare them?

Model a realistic month. Take your expected connected minutes, then for each vendor add the platform fee, every pass-through component at the models you would actually use, and telephony if it is not included. Compare those totals. A pass-through model can beat a bundled one at volume if you tune the stack — and can quietly exceed it if you do not.

What else affects the real cost?

Connect rate, above all. You pay for conversation minutes, but the campaign’s economics depend on cost per useful outcome, and a list with a 5% connect rate costs twenty times as much per conversation as one at 100%. Average call length matters similarly: a 90-second qualifying call and a six-minute support call have very different unit economics at the same per-minute rate.

Frequently asked questions

What is the cheapest voice AI platform?: On published headline rate, Vapi at 5¢ per minute, then Bolna at 6¢. But Vapi’s figure excludes speech and model costs, which are passed through, so the cheapest headline is not reliably the cheapest bill. Model your own volume and model choices across vendors before deciding.
Are phone numbers included in these prices?: It varies, and it is often the component people forget. Rexa.ai includes phone numbers in its per-minute rate by default, provisioning the number itself, and also supports bringing your own carrier account — on that path your carrier bills you directly. Bolna includes a free number for the first 30 days on its Pilots plan. Deepgram does not provide telephony at all. Vapi’s pricing page prices concurrency per line but does not state number provisioning.
Is a voice AI agent cheaper than a human caller?: Per minute, substantially — but that comparison flatters the agent, because a person handles complexity an agent escalates. The honest comparison is cost per resolved outcome across the whole queue, counting the calls that still reach a human. For high-volume repetitive calling the gap is large; for complex or high-value conversations it narrows quickly.

Scope, limitations, and sources

All vendor figures below were read from each vendor’s own pricing page on the access dates shown and are list prices for self-serve or entry plans. Committed-volume and enterprise pricing is negotiated and not public. Pricing in this category changes frequently — re-read the sources before relying on any figure. Rexa.ai publishes this page and is one of the vendors compared; the figures are cited so you can check them rather than take our word for it.

Sources

Vapi Build-plan rate of $0.05/min and the statement that model costs are passed on to the customer. vapi.ai (accessed August 8, 2026)
Bolna per-minute rate of 6.00¢ with volume discounts and BYOK support. bolna.ai (accessed August 8, 2026)
Deepgram Voice Agent API rates ($0.075/min standard, $0.163/min advanced) and that Deepgram does not provide telephony. deepgram.com (accessed August 8, 2026)
Synthflow enterprise contracts starting at $30,000 annually with custom-scoped pricing. synthflow.ai (accessed August 8, 2026)
Rexa.ai rate of 10¢ per minute and the components it includes. rexa.ai (accessed August 8, 2026)

Spotted something out of date or wrong? Tell us on the contact page and we will correct it and update the review date.