AI Voice Agent Cost per Minute in 2026: Vapi vs Retell vs ElevenLabs vs Bland, Line by Line

On October 2026 list prices, running an AI voice agent costs about $0.09–0.23 per minute once every line is counted: the platform, speech-to-text, the language model, the voice and the phone carrier. Vapi starts at $0.05 for hosting alone, Retell at $0.07, ElevenLabs at $0.08 and Bland at $0.12. Those headline rates don’t cover the same things.

This guide splits the minute into its parts, prices four platforms at four monthly volumes and lists the charges that sit outside the per-minute rate. Every figure was read on the vendor’s own pricing page on 7 October 2026. Rates in this market change often, so recheck before you sign. We build AI voice agents and AI receptionists and run one on our own studio number, so we have a stake here. In our projects usage is billed at cost on the client’s accounts, which is why the estimate has to be right. Build prices and care plans are covered in how much an AI agent costs; this article is only about the running cost of calls.

AI voice agent pricing at a glance (October 2026)

The headline numbers answer different questions. Vapi’s $0.05 is the fee for running the conversation. Its own calculator, set to 1,000 minutes with Deepgram, OpenAI and ElevenLabs, shows an estimated total of $82–129, which is $0.08–0.13 a minute. Retell’s range already includes a voice and a model. ElevenLabs includes everything except the model. Bland includes everything except the carrier. Compare totals for your configuration, not headlines.

AI voice agent cost per minute by component

A phone agent is a chain of five services. Each one bills by the minute or converts to it.

Three things stand out:

• The model is the widest range. On Retell’s card the dearest model costs 200 times the cheapest. Retell marks GPT 5.6 Terra and Claude 5 Sonnet, both $0.064, as recommended; smaller models such as Claude 4.5 Haiku cost less than half of that.
• The voice can cost more than the model. A premium ElevenLabs voice on Retell is $0.04–0.10 a minute against $0.015 for a platform voice.
• Telephony is the smallest line. In the US it stays under two cents a minute. Toll-free numbers and international routes cost more.

Two Retell configurations, added up

• Lean: voice infrastructure $0.055 + platform voice $0.015 + Claude 4.5 Haiku $0.025 + Retell telephony $0.015 = $0.11 per minute.
• Premium: voice infrastructure $0.055 + ElevenLabs v4 Turbo voice $0.08 + GPT 5.4 $0.08 + Retell telephony $0.015 = $0.23 per minute.

It is the same platform and the same call at twice the price. Choose the voice and model after test calls with real callers’ questions, not from a price list.

Monthly bills at 500, 1,500, 5,000 and 20,000 minutes

The volumes are illustrative. 500 minutes is about 170 three-minute calls a month. 1,500 minutes is 500 such calls, the volume we used in the cost guide.

How each column is built:

• Vapi. $0.0818–0.1289 a minute from the calculator’s default stack: hosting $0.05, Deepgram $0.0095–0.0099, OpenAI $0.0077–0.0452, ElevenLabs $0.0146–0.0238, on Vapi’s free telephony. Your own Twilio number adds $0.008 a minute for inbound calls, which is $12 at 1,500 minutes.
• Retell. The two configurations above, with Retell telephony. Add $2 a month per number.
• ElevenLabs. The plans are prepaid bundles at one rate: Creator is $22 for 275 minutes, Pro $99 for 1,238, Scale $299 for 3,738 and Business $990 for 12,375, and extra minutes cost $0.08. So 1,500 minutes on Creator is $22 + 1,225 × $0.08 = $120. The model is billed on top by usage, and the page gives no flat per-minute figure for it.
• Bland. Start is capped at 100 calls a day. At three minutes a call, 20,000 minutes is more than twice that, so it needs Build: $299 + 20,000 × $0.12 = $2,699. On price alone Start stays cheaper up to about 14,950 minutes a month ($299 ÷ $0.02). Telephony is extra on both plans.

What the table shows:

• At low volume, usage is small next to the build. 500 minutes costs $40–115 a month on any of these platforms.
• The configuration matters more than the platform. The $0.12 a minute between Retell lean and Retell premium is wider than the gap between any other two columns in the table.
• At high volume, small differences add up. At 20,000 minutes, the $0.15 between an $0.08 and a $0.23 setup is $3,000 a month.

What our own phone line costs

Our studio number is answered by an AI agent built on ElevenLabs Agents, with a Kyiv number connected over SIP and Claude Haiku 4.5 as the model. From our setup notes of 2 October 2026:

• the number costs about $2 a month plus a $1 connection fee, and inbound calls on it are free;
• the platform charges $0.08 a minute, on the $22 Creator plan;
• after each call a larger Claude model writes a brief and a cost estimate from the transcript, at about $0.05–0.15 per call;
• a five-minute call comes to about $0.50 all in.

We also set limits: 3 concurrent calls, 60 calls a day, 15 minutes per call, and the agent hangs up after 30 seconds of silence. Limits like these are a cost control as much as a safety one. The line went live in October 2026, so treat these as rate-card figures, not a year of invoices.

Costs outside the per-minute rate

• Concurrency. Vapi includes 4 concurrent calls and sells extra lines at $10 each a month. Retell includes 20 and charges $8 for each extra. ElevenLabs bills minutes above your plan’s limit at the $0.16 burst rate. Bland caps Start at 100 calls a day and Build at 2,000.
• Silence and hold time. Retell’s FAQ states that billing covers the whole call, silence included, because speech-to-text keeps listening. Retell counts calls to the nearest second.
• Transfers. On Retell the agent fee stops after a transfer and only telephony continues. Bland charges $0.05 (Start) or $0.04 (Build) per transferred minute on its own telephony and nothing extra if you bring your own carrier.
• Numbers and texts. A US local number is $1.15 a month at Twilio and a toll-free one $2.15; a Retell number is $2. Retell charges $0.01 per SMS. Business texting to US numbers also needs carrier registration (A2P 10DLC), with its own fees, and texts should go only to people who agreed to receive them.
• Add-ons. On Retell: knowledge base +$0.005 a minute, noise removal +$0.005, guardrails +$0.005, removal of personal data from transcripts +$0.01, automated quality review $0.10 a minute after 100 free minutes. Vapi’s HIPAA add-on is $2,000 a month.
• Outbound extras. Twilio charges $0.0075 per call for answering-machine detection. Retell charges $0.005 per dial for batch calls and $0.10 per outbound call for a branded caller ID. Calls that reach voicemail still use minutes.
• Recording. Twilio lists $0.0025 a minute to record and $0.0005 a minute per month to store.
• Work after the call. Summaries, CRM updates and call scoring are extra model calls. On our line that is the $0.05–0.15 per call above.
• Test calls. They are billed like real ones. A test set of 100 three-minute calls is 300 minutes before launch, and you repeat part of it after every change.
• Support plans. Vapi’s Core package is $29 a month. Its Pro package has a $999 monthly minimum.
• Markups. If the accounts belong to an agency, ask for the platform invoices. One client post in our September 2026 research said plainly that hidden monthly markups were not acceptable.

Platform, all-in rate or custom stack: decision rules

These are our rules of thumb, not vendor guidance.

• Pick an all-in rate (Bland) when you want one number per minute and stable invoices, and you can live with the vendor’s choice of model and voice.
• Pick ElevenLabs Agents when voice quality is the priority and the call flow is simple. The fixed part is flat at $0.08, and only the model varies.
• Pick Retell when you want one invoice but your own choice of model and voice, or you expect bursts: 20 concurrent calls are included.
• Pick Vapi when you have a developer and want to control each provider separately. You pay for that control with more accounts and more lines on the bill.
• Pick a subscription receptionist when your calls are standard. One done-for-you service we reviewed in September 2026 charged $99–149 a month plus a $1,500 setup fee, with the lower tier capped at 100 calls.
• Build your own stack only at volume. Going direct to a carrier, a speech-to-text provider, a model and a voice removes the platform fee of $0.05–0.055 a minute. At 1,500 minutes that saves $75–83 a month, less than an engineer’s hour. At 20,000 minutes it is $1,000–1,100 a month, which can pay for building and maintaining turn-taking, interruption handling and monitoring yourself.

If the calls are simple enough for a menu, you may not need a voice agent at all. We compare the options in AI agent vs chatbot.

What to check before you commit

Typical mistakes we see in estimates: counting only connected talk time, forgetting the second and third concurrent line on Monday mornings, pricing the demo model and then switching to a larger one, and leaving out test minutes.

How we would approach it

Hold any vendor, including us, to the checklist above. Our process:

1. Brief. You send your call volume, busiest hours and what callers ask. Within 24 hours you get a call flow, a stack and a fixed quote.
2. Usage estimate in writing. Minutes, concurrency and the per-minute rate for the chosen model and voice, line by line, before the build starts.
3. Demo call on your data before the main invoice, so the model and voice are chosen by ear.
4. Your accounts. The carrier, the voice platform and the model are in your name, and usage is billed at cost.
5. Limits and alerts on call length and daily volume from day one.

An AI receptionist build starts at $2,000 and usually goes live in 1–2 weeks; a Growth build with chat channels and CRM starts at $3,500 and takes 2–3 weeks. Thirty days of tuning are included, and care plans start at $250 a month. If you already have a Vapi or Retell agent that costs more than it should, a rescue audit starts at $400. Send your numbers through the voice agents page and you’ll get the estimate with the quote.