Voice AI pricing comparison: calculate the complete call cost

The most useful voice AI price is the cost of completing your intended task. A platform-only minute rate, a bundled AI minute and a carrier minute measure different things. Put them on the same worksheet before deciding which service is less expensive.

Public prices and account terms can change. This guide explains how to compare a current quote; it does not present an invented universal ranking.

Understand the pricing model first

Use each provider's own current information: Vapi pricing, Retell pricing, Bland pricing, and Burki pricing. A quote for one specific configuration is more useful than a headline copied into a comparison table.

Build a monthly worksheet

For each candidate, enter these quantities and their billing units:

• Connected call duration, split by inbound and outbound destination.
• Model input and output, including audio or text units where applicable.
• Speech generation and transcription if separately charged.
• Number rental, recording/storage, and selected support or compliance terms.
• Transfer-leg duration, especially when two telephone legs overlap.
• Fixed monthly fees and minimum commitments.

Do not add separate STT and TTS estimates to a bundled speech-to-speech quote unless those components are actually billed separately. Conversely, a BYO provider bill remains part of your cost even if the platform never collects it.

Worked example: why the headline can mislead

The following numbers are hypothetical, not vendor rates. Suppose an assistant handles 1,000 calls averaging three connected minutes. Configuration A has an effective variable cost of $0.12 per minute and $20 in fixed monthly costs. Configuration B costs $0.16 per minute with $10 fixed.

A costs $380 and B costs $490. But if A completes only 700 intended tasks while B completes 950, their costs per successful task are about $0.54 and $0.52 respectively. The higher minute rate can produce the lower cost per useful result.

This calculation also needs the cost of human recovery where failures create extra work. Treat automated outcome labels as evidence to review, not an unquestioned denominator.

Reconcile a pilot invoice

Select a few completed calls and compare the estimated charge, any reserved balance, and the settled charge. A reservation is a temporary exposure limit, not a second purchase. Check the provider and carrier statements for the same period, including short failed calls and number rental.

If a call was transferred, inspect every leg. If it failed before a useful conversation, check whether connection time or provider work was still billable. These are accounting questions, not proof of a billing error by themselves.

What to optimize first

Remove unused recurring resources, shorten unnecessary assistant monologues, and fix retries that repeat expensive work. Then compare model or voice changes using the same scenarios. Reducing answer quality to save fractions of a cent can increase retries and abandonment.

For a more detailed distinction between managed and direct provider bills, read BYO versus managed usage. Start the purchase decision with the current price page, not an old screenshot or a platform-only comparison.