AI Voice Agent Cost Per Minute in 2026: Vapi, Retell, Bland and ElevenLabs, Decoded

Technology
Sleek microphone on a dark studio desk surrounded by floating translucent audio waveform layers in blue light
November 8, 2026 Updated Nov 16, 2026 9 min read 78 views

AI Voice Agent Cost Per Minute in 2026: Vapi, Retell, Bland and ElevenLabs, Decoded

How much does an AI voice agent cost per minute is the first question every voice-AI evaluation starts with, and the first question the pricing pages answer badly. The headline rates you see advertised, $0.05 per minute on Vapi and $0.055 on Retell, are platform fees: the phone call that actually reaches a customer costs roughly double once speech recognition, the language model, the voice and the phone line are assembled. Reading the vendors' own pricing pages this month, a realistic all-in minute in 2026 lands between $0.07 and $0.31 depending on which components you choose, and the same 1,000-minute month can cost $82 or $184 across four mainstream platforms. This guide decomposes the minute into its five billable parts, prices each platform honestly, and shows you where the real cost decisions hide.

The Five Parts of One Voice-Agent Minute

Every phone-based AI agent bills the same call in layers, and most sticker shock comes from teams comparing one layer against another vendor's bundle:

  • Platform/orchestration fee: the vendor's own charge for running the agent loop. Vapi lists $0.05 per minute for hosting; Retell's voice infrastructure is $0.055 per minute (both vendors' pricing pages, checked November 2026).
  • Speech-to-text: transcribing the caller. Often bundled into the infrastructure rate, sometimes billed separately by provider (Deepgram and similar).
  • The LLM: the largest variable. Retell publishes per-minute model pricing from $0.0016 per minute (GPT-5 nano) to $0.384 (GPT-6.1 Sol Ultrafast), with mid-size models like Claude 5 Sonnet at $0.064 and GPT-5 at $0.04.
  • Text-to-speech: the voice you hear. Retell's platform voices run $0.015 per minute, ElevenLabs Flash on top of Retell is $0.040, and ElevenLabs v3/v4 premium voices reach $0.10 per minute.
  • Telephony: the phone leg. Retell charges $0.015 per minute on a provided US number plus $2 per number per month; Vapi passes Twilio/Telnyx through at cost or runs free on your own SIP.

Retell's own cost calculator makes the arithmetic explicit: a default configuration assembles to $0.11 per minute, $0.04 LLM + $0.055 voice infrastructure + $0.015 TTS + telephony. The vendor's advertised floor of $0.07 assumes the cheapest components; the $0.31 ceiling assumes a frontier model and a premium voice. Both numbers are true. Neither is your bill until you pick components.

What the Four Mainstream Platforms Actually Charge

All four figures below come from the vendors' own pricing pages this month:

  • Vapi: $0.05/min hosting, everything else at cost. Vapi's own calculator estimates 1,000 minutes at $82-$129/month on its default stack (Deepgram transcriber, OpenAI model at $8-$45 per 1,000 minutes, ElevenLabs voice at $15-$24). Add-ons are subscriptions, not minutes: the Core success package is $29/month for 10 concurrent calls, and Pro starts at a $999/month minimum with 30 concurrent calls and an uptime SLA.
  • Retell AI: $0.07-$0.31/min all-in, pay-as-you-go, $10 free credit. Component-priced on one bill; 20 concurrent calls included, then $8 per concurrency per month. It bills to the second and stops the agent fee once a call transfers to a human, a detail that materially changes call-center math.
  • Bland AI: the bundled option: $0.14 per minute on the free Start plan with LLM, speech-to-text and TTS all included (telephony pass-through), or $0.12 plus a $299 monthly fee on Build with 50 concurrent calls. One number, no component spreadsheet.
  • ElevenLabs Agents: plan-based minutes with $0.08 per minute after included allowances, LLM billed separately at provider rates, and bring-your-own telephony via Twilio or SIP with no platform phone fee.

Independent audits that assembled every component reached consistent all-in ranges: about $0.11-$0.135 per minute on Vapi and Retell with default mid-size stacks, $0.11-$0.14+ on Bland, and per independent comparisons published in September-October 2026 (stackbinary.io, aliteq.com, superdupr.com, each stating they read the vendor pages directly), a 1,200-minute month lands roughly $111-$184 across the four. Bland's own FAQ makes the sharpest competitive point: competitor advertised rates are "the platform fee only," with production stacks realistically at $0.13-$0.30 per minute.

Worked Monthly Bills: 1,000 vs 10,000 Minutes

Volume changes the winner, so price both scenarios before signing anything (per-minute rates above, plus platform surcharges):

  • 1,000 minutes/month (~330 two-minute calls): roughly $80-$130 on Vapi or Retell pay-as-you-go with a mid-size model, $140 on Bland Start, ~$80-$110 on ElevenLabs plus the LLM. At this scale all four are cheaper than a single hour of human phone time: the decision should be made on features and concurrency, not price.
  • 10,000 minutes/month: the independent platform comparisons converge on $780-$1,300 depending on model and voice choices. This is where Vapi's bring-your-own-keys advantage appears: if you already hold a model contract, Vapi bills nothing on that line, and it is the only route that gets materially under $0.10 per minute at scale.
  • Concurrency is the hidden tier gate. Retell includes 20 simultaneous calls; Vapi gives 4 free and 10 at $29/month; Bland caps 10 on Start and 50 on Build. A call spike that forces a tier upgrade can cost more than the minutes themselves: size concurrency to your busiest hour, not your average.

A useful sanity anchor from the enterprise side: analyst projections circulating in 2026 vendor materials claim agentic AI will cut contact-center labor costs by tens of billions of dollars this year with roughly one in ten interactions automated. Treat vendor-sourced macro numbers as direction, not data: the per-minute math above is the part you can verify on your own invoice.

What Actually Moves Your Bill (and What Does Not)

  • The model choice is the biggest dial. Swapping a frontier model for a mid-size or nano-tier model is a 10-100x difference on the LLM line. Most receptionist workloads (hours, bookings, qualification) do not need GPT-class reasoning every turn; route simple turns to a cheap model and escalate hard questions.
  • Premium voices are a luxury tax. $0.015 versus $0.10 per minute is a 6x spread on TTS alone. Test whether callers can tell before paying for ElevenLabs v4.
  • Call transfers should stop the meter. Retell stops billing the agent fee on transfer; platforms that keep billing during the human leg quietly inflate call-center bills.
  • Zero data retention and HIPAA are paid add-ons. Vapi lists HIPAA compliance at roughly $2,000 per month and zero-data-retention at $1,000 per month (per its pricing-page terms cited in September 2026 comparisons): compliance is a line item, not a checkbox.
  • Build cost is separate from run cost. Per-minute pricing is the operating bill; the development or agency build is a one-time project typically quoted in days-to-weeks of engineering, which is why many businesses buy a managed receptionist product at $99-$600/month instead of running raw platforms.

FAQ

How much does an AI voice agent cost per minute, realistically?

About $0.08-$0.15 per minute all-in for a US phone call on a mid-size model as of November 2026, assembling platform fee, speech-to-text, LLM, text-to-speech and telephony. Advertised floors of $0.05-$0.055 are hosting fees only; premium models and voices push the same minute toward $0.20-$0.31. Every figure above is traceable to the vendors' own pricing pages.

Is Vapi really $0.05 a minute?

No: $0.05 is Vapi's hosting fee. Speech-to-text, the language model, the voice and the phone line are billed separately at provider cost, which Vapi passes through without markup. Vapi's own calculator puts a default 1,000-minute month at $82-$129, which works out to about $0.08-$0.13 per minute. Bringing your own model keys is the legitimate way to lower it further.

Which is cheaper, Retell or Bland?

It depends on your component choices and volume. Retell's component pricing (~$0.11 assembled with mid-tier defaults) beats Bland's flat $0.14 on Start if you choose efficient models and voices, and its $8/line concurrency add-on is cheaper than Bland's plan jump to Build ($299/month). Bland wins on simplicity: one number, no spreadsheet, LLM included: predictable bills are worth money too.

How much does it cost to build a custom AI voice agent?

Operating costs follow the per-minute math above; the build itself is the other bill. DIY on a platform costs weeks of developer time; a done-for-you receptionist product typically runs $99-$600 per month all-inclusive, and a custom agency build runs into four figures one-time. If your call flows are standard (answer, qualify, book), a product beats a project. Our guide to AI agent build-vs-buy costs breaks down where each route pays.

Do per-minute prices include HIPAA compliance?

Generally no. ElevenLabs, Retell and Bland offer BAAs on enterprise plans, while Vapi prices HIPAA support as a roughly $2,000/month add-on (per September 2026 pricing-page comparisons: verify current terms directly, as these change). For any healthcare voice deployment, ask for the BAA document and the compliance price before the per-minute rate.

ArcBeta designs and operates production voice agents (component selection, latency tuning, transfer logic and compliance hosting included) for appointment-driven businesses. Start a voice-agent scoping call with us. To see what this per-minute math looks like in a specific vertical, read our breakdown of AI receptionist costs and options for small business.