What a voice agent really costs per minute
Five components, each priced separately by a different vendor. Here is the formula and the current published prices, so you can do the arithmetic yourself.
Want to see it first? Book a demo
- The full formula
- Dated vendor prices
- No email required
The short answer
- The formula: per-minute cost = platform or hosting fee + LLM tokens + text to speech + speech to text + telephony. Every one is billed by someone, and a headline price that quotes only the first is quoting a fraction.
- Using published prices as of September 2026, a typical configuration lands between roughly $0.07 and $0.31 per minute depending on which model and which voice you pick. The model choice is usually the biggest single lever.
- Three numbers of your own finish the sum: your answer rate, your average call length and the share of calls that book. Those decide the economics far more than a cent per minute, so read them off your call log before you compare vendors.
The formula, step by step
The formula
Cost per minute = platform fee + LLM + TTS + STT + telephony
Then, to get to what you actually care about:
Cost per call = cost per minute × average call length in minutes
Cost per booked appointment = cost per call ÷ the share of calls that end in a booking
The second line is the one that decides whether this is worth doing. A call costing thirty cents that books nothing is expensive. A call costing a dollar that books a job is not.
The five components, with published prices
Each figure is from the vendor's own pricing page. Check them yourself before you commit to a budget.
| Component | What it is | Published price | Source |
|---|---|---|---|
| Platform, Vapi | Orchestration and hosting per minute of call | $0.05 per minute on the Build plan; volume pricing on Scale | vapi.ai/pricing |
| Platform, Retell | Their voice infrastructure per minute | $0.055 per minute | retellai.com/pricing |
| LLM | The model that decides what to say, billed on tokens in and out | Retell lists model options from $0.045 to $0.16 per minute depending on the model | retellai.com/pricing |
| Text to speech | Turning the reply into audio | $0.015 per minute on standard voices; $0.040 per minute on ElevenLabs voices, per Retell's list | retellai.com/pricing |
| Speech to text | Turning the caller's audio into text | Billed at provider cost by Vapi, and $0 if you bring your own key | vapi.ai/pricing |
| Telephony | The actual phone call and the number | $0.015 per minute in Retell's United States Twilio example | retellai.com/pricing |
Prices read from each vendor's own pricing page in September 2026. Vendors change these; treat the numbers as a starting point and verify before budgeting.
A worked example
Retell publishes an example that totals $0.11 per minute: $0.04 for the LLM, $0.055 for their voice infrastructure and $0.015 for text to speech. Add United States telephony at $0.015 and you are at roughly $0.125 per minute before anything else.
Retell's own stated range across configurations is $0.07 to $0.31 per minute. The spread is almost entirely the model and the voice: a premium model and premium voices roughly triples the bill.
Now apply the second and third lines of the formula. At $0.125 a minute and a four-minute average call, a call costs about fifty cents. If one call in eight ends in a booking, each booking costs about four dollars in agent time. Whether that is good depends entirely on what a booking is worth to you, which is a number only you have.
The four costs the formula misses
- Failed and abandoned calls. You pay for the seconds even when the caller hangs up at hello, and those calls do not appear in a per-booking number unless you count them.
- Phone numbers. A monthly rental per number, separate from per-minute usage.
- Carrier registration. Business texting in the United States needs registration, which has its own fees and its own delay.
- Your time building it. The prompt, the testing, the escalation rules and the week of listening to recordings. This is usually the largest real cost in month one and it never appears on a pricing page.
Where SalesCrew sits in this
SalesCrew places outbound calls through your own Vapi account, so the per-minute cost above is yours directly and there is no markup in between. Inbound works the same way: the AI receptionist answers on your own Vapi or Retell account, books the appointment or transfers the call, and the call log and voicemail transcript sit in the same thread list as your email.
Model spend inside SalesCrew for drafts, summaries and scoring is separate from voice, metered against your allowance, with a soft alert at eighty percent and a hard stop at a hundred that applies to AI only. Bringing your own model key keeps that spend on your account entirely.
Work out your own number before anyone quotes you one
Five components, four hidden costs, and a per-booking figure that only you can complete.
Want to see it first? Book a demo