Cost per minute

In one sentence

Cost per minute is the unit price of voice processing, covering speech recognition, language model inference, speech synthesis and, in telephony, carrier charges, and it is how voice platforms bill and how voice unit economics are modeled.

USD 0.05 to 0.09 per min Vendor published pricing (Vapi, Retell, Bland), 2026Last reviewed 4 August 2026

Definition

Cost per minute is what it costs to run a conversation for one minute.

Unlike normal software, it costs you something every single time someone uses it, and the advertised price is usually just the platform cut rather than the whole bill.

Cost per minute is the operative unit because voice cost genuinely scales with duration, unlike most software. Read the stack correctly and pricing, margin and the judgement on which segments are servable all follow from it.

The stack that produces the figure

  • Speech recognition, charged per minute or per hour of audio.
  • Language model inference, charged per token, which varies with prompt size and response length rather than with time directly.
  • Speech synthesis, charged per character or per minute of generated audio.
  • Telephony carrier charges, which apply to phone deployments and not to browser ones.
  • Infrastructure: media handling, retrieval, storage, logging.

Why the browser case is structurally cheaper

  • Removing telephony removes an entire cost line.
  • There is no number provisioning, no carrier relationship and no per-minute termination charge.
  • This is a genuine structural difference rather than a pricing choice.

The controllable drivers

  • Conversation duration, which multiplies almost every component.
  • Prompt size, which drives inference cost and is frequently bloated invisibly.
  • Model selection, where the spread is large on both cost and latency.
  • Retrieval efficiency, where fewer, better passages reduce cost and improve quality at the same time.

Why headline rates mislead

  • Published platform rates are base rates covering the platform margin, not the underlying model and telephony costs.
  • All-in production cost lands materially above the headline.
  • Comparing platforms on headline rate without normalizing for what is included produces incorrect conclusions.

Common misconception

That cost per minute will fall to negligible levels as models get cheaper. Per-unit costs have fallen substantially and usage has grown faster. The variable cost structure is durable, and any business betting that it vanishes is standing on ground that has not held so far.

Why it matters commercially

Cost per minute is the denominator in voice unit economics and the number that determines which customer segments are servable at what price. Get the stack wrong and every pricing and margin decision downstream inherits the error.

In voice specifically

Voice bills per minute because its cost scales with how long someone talks. Text does not: a chat is charged per token or per message and carries no recognition or synthesis line. That is why per-minute is the operative unit for voice and never becomes the unit for text.

Where AsqVox fits

A browser-based widget carries no telephony line in its cost stack, which removes a component telephony voice agents must absorb. The same recognition, model and synthesis pipeline runs on both sides; browser voice simply leaves the carrier band an empty outline.

Visual

What a minute actually costs

What a minute actually costsTelephony carrierphone deployments only; a browser widget drops this band entirelyInfrastructuremedia handling, retrieval, storage and loggingSpeech synthesischarged per character or per minute of generated audioModel inferencecharged per token, driven by prompt size and response length rather than by timeSpeech recognitioncharged per minute or per hour of audio processedAll-in production cost, roughly USD 0.11 to USD 0.33 per minute, is the sum of everything above, and it sits above every headline platform rate.Browser voice against telephony voiceSame recognition,model and synthesispipeline on bothsides. Telephony addsthe carrier band ontop; browser voiceleaves it an emptyoutline. The samestack, one fewer costline.Headline platform rates (Vapi ~USD 0.05, Retell ~USD 0.07, Bland ~USD 0.09) sit well below the full stack. They cover the platform margin, not the model and telephony passed through beneath.

Voice costs money every time it is used. That does not stop being true.

The layers name what each line is charged for; the source publishes the all-in range (USD 0.11 to 0.33) and the platform base rates, not per-component dollar figures, so no layer carries a measured amount. The telephony carrier band is the line a browser widget removes, which is a structural saving rather than a discount. In India, Bolna prices from around Rs 7 per minute falling to around Rs 3 at volume.

Statistics

Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.

Platform base rates are around USD 0.05 per minute for Vapi, USD 0.07 for Retell and USD 0.09 for Bland.

USD 0.05 to 0.09 per minvendor claim

Vendor published pricing (Vapi, Retell, Bland), 2026 - Base rates covering the platform margin, not the underlying model and telephony costs. The full stack sits above them.

All-in production cost, once recognition, model, synthesis and telephony are stacked, lands around USD 0.11 to USD 0.33 per minute.

USD 0.11 to 0.33 per minindustry range

Industry-reported production cost range, 2026 - The delivered cost. Where in the range a deployment lands is driven mostly by conversation duration and prompt size.

In India, Bolna prices from around Rs 7 per minute falling to around Rs 3 per minute at volume, following a USD 6.3 million seed round led by General Catalyst in January 2026.

Rs 7 to Rs 3 per minvendor claim

Bolna published pricing and funding announcement, 2026 - A volume-tiered range. Vendor pricing changes often, so any published rate should be dated and linked to a current pricing page.

A human-handled call costs roughly USD 7 to USD 12 against roughly USD 0.40 for an agent-handled call.

USD 7 to 12 vs USD 0.40industry range

Widely cited industry range, 2025 - An industry range rather than an audited figure, varying substantially by geography and complexity. It is the comparison basis a per-minute figure is usually converted into.

Provider pricing changes frequently, so any published rate should be dated and linked to the current pricing page.

-no reliable figure

A standing caveat rather than a figure. A rate quoted with no date has an unknown shelf life.

Audio token pricing in speech-to-speech architectures follows different schemes and is not directly comparable to text token pricing.

-no reliable figure

Which is why a like-for-like per-minute comparison across architectures is harder than it looks. Normalize for what is being charged before comparing.

There is no published benchmark for average conversation duration in website voice contexts, which is the input required to convert per-minute cost into per-conversation cost.

-no reliable figure

Duration multiplies almost every component, so this gap matters more than any single rate. A deployment has to measure its own.

Examples

In practice

A team selects a platform on headline per-minute rate and finds actual costs materially higher, because the headline covers the platform layer while recognition, model and synthesis are passed through. Rebuilding the comparison on all-in cost per minute across representative traffic reverses the ranking.

The everyday version

Cost per minute is what it costs to run a conversation. Unlike normal software, it costs you something every single time someone uses it, and the advertised price is usually just the platform cut rather than the whole bill.

Usage

Who says it

  • Platform operators, finance teams and investors, in unit economics and margin analysis.
  • Buyers, in pricing comparison, where the headline rate is the number most likely to be quoted.

Where it turns up

  • In pricing pages, cost models and unit economics analysis.
  • In vendor selection, through questions about cost predictability, rate limiting and overage.

Common misuse

  • Comparing headline rates without normalizing for what is included, which produces incorrect conclusions.
  • Assuming costs trend toward zero, when usage has historically grown faster than unit costs have fallen.
  • Ignoring prompt size and retrieval volume as cost drivers, which are invisible on a bill and substantial in aggregate.

Questions people ask

How much does a voice agent cost per minute?

All-in production cost lands around USD 0.11 to USD 0.33 per minute, once speech recognition, the model, synthesis and any telephony are stacked. Platform base rates are lower, around USD 0.05 to USD 0.09 per minute for Vapi, Retell and Bland, but those cover the platform margin only. In India, Bolna prices from around Rs 7 per minute falling to around Rs 3 at volume.

Why is the platform rate lower than my actual cost?

Because the published platform rate is a base rate covering the platform own margin, not the underlying model and telephony costs passed through beneath it. All-in production cost lands materially above the headline. Comparing platforms on headline rate without normalizing for what is included reverses rankings once real traffic is measured.

Is browser voice cheaper than phone voice?

Structurally, yes. A browser deployment carries no telephony carrier charge, no number provisioning and no carrier relationship, which removes an entire cost line. The recognition, model and synthesis pipeline is the same on both sides, so the difference is one fewer band on the stack rather than a pricing choice.

Will voice cost per minute fall to near zero?

No. Per-unit costs have fallen substantially, but usage has grown faster, and the variable cost structure is durable. Voice costs money every time it is used, and that does not stop being true. The levers you actually control are conversation duration, prompt size, model selection and retrieval efficiency.

Share this definition

Last reviewed 4 August 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.