What one minute of a voice agent costs
A voice agent minute is built from four vendors: the voice platform (speech-to-text, turn-taking, hosting), telephony, text-to-speech and the language model. The voice and the model you choose decide most of the difference. Prices below are Retell AI's published per-minute rates for a US call, checked 22 November 2026.
| Stack | Platform | Telephony | Voice | Model | Total |
|---|---|---|---|---|---|
| Lean | $0.055 | $0.015 | $0.015 standard voice | $0.032 GPT-6.1 Terra | $0.117 |
| Typical | $0.055 | $0.015 | $0.040 ElevenLabs Flash | $0.048 Gemini 3.8 Flash | $0.158 |
| Premium | $0.055 | $0.015 | $0.100 ElevenLabs v4 | $0.064 Claude Sonnet 5.5 | $0.234 |
Independent comparisons put typical all-in costs at $0.08–0.13 a minute. Negotiated rates, other platforms (Vapi, Bland, your own stack) and non-US telephony change the numbers, so enter your own cost in the calculator.
Margin depends on how much of the bundle customers use
Example plan: $299 a month for 2,000 minutes ($0.149 per included minute), extra minutes at $0.20. The price is fixed; the cost grows with every minute used.
| Stack and billing | 25% used | 50% used | 75% used | 100% used | Break-even |
|---|---|---|---|---|---|
| Lean, monthly | 80% | 61% | 41% | 22% | Never |
| Lean, annual (−20%) | 76% | 51% | 27% | 2% | Never |
| Typical, monthly | 74% | 47% | 21% | −6% | 95% |
| Typical, annual (−20%) | 67% | 34% | 1% | −32% | 76% |
| Premium, monthly | 61% | 22% | −17% | −57% | 64% |
| Premium, annual (−20%) | 51% | 2% | −47% | −96% | 51% |
Break-even is the share of included minutes at which delivery cost equals what the customer pays.
Four things that quietly turn a plan negative
Premium voices on every plan
A premium voice costs up to $0.100 a minute against $0.015 for a standard one. If any customer can pick it at no extra charge, the example plan goes from profitable at any usage on a lean stack to losing money at 64% of the bundle on a premium one.
Annual discounts on unchanged bundles
A 20% discount removes 20% of revenue and none of the cost. On a typical stack, a fully used annual plan runs at −32%. Offer fewer minutes on annual plans, or a cap.
Overage priced like included minutes
At $0.20 a minute, extra minutes earn 42% on a lean stack, 21% on a typical one and −17% on a premium one. Heavy users never earn the plan back.
Billed time nobody counts
Hold and silence are billed. Short calls with a dynamic greeting bill as 10 seconds, prompts over 4,000 tokens bill as more minutes, and add-ons cost $0.005–0.01 a minute each.
The floor price for a minute
To earn a target margin on every used minute, price it at cost per minute ÷ (1 − target margin). Use this for overage, and as a check on the price per included minute your heaviest customers actually pay.
| Stack | 50% margin | 60% margin | 70% margin |
|---|---|---|---|
| Lean ($0.117) | $0.234 | $0.292 | $0.390 |
| Typical ($0.158) | $0.316 | $0.395 | $0.527 |
| Premium ($0.234) | $0.468 | $0.585 | $0.780 |
An average cost per minute hides which customers lose money
This page uses one cost per minute for everyone. In practice it varies by customer: their voice, call length, prompt size and hold time. Ganivra records the cost of every call from every vendor and shows margin per named customer, so you can see which accounts are negative and why.
Questions
What does a voice AI agent cost per minute?
On Retell's published prices (checked 22 November 2026), a US call costs about $0.117 a minute with a standard voice and a small model, $0.158 with ElevenLabs Flash and Gemini 3.8 Flash, and $0.234 with a premium ElevenLabs voice and Claude Sonnet 5.5. That covers the platform, telephony, voice and model. Retell quotes $0.07–0.31 a minute overall.
What gross margin should a voice AI company target?
Software companies usually aim for 70–80% gross margin. Voice AI companies selling minutes rarely reach that, because every minute has a real vendor cost. 50–60% on a fully used plan is a common working target; the important part is that the margin stays positive for your heaviest customers, not just on average.
How do I calculate break-even usage on a plan with included minutes?
Divide the monthly price by the cost of the included minutes. A $299 plan with 2,000 minutes at $0.158 a minute costs $316 if fully used, so it breaks even at 299 ÷ 316 = 95% usage. If the customer goes over the bundle, the overage price must be above your cost per minute or the plan never recovers.
Does an annual discount hurt voice AI margins?
It cuts revenue while the cost of each minute stays the same. On the example plan with a typical stack, a 20% annual discount moves the margin of a customer using every included minute from −6% to −32%. Annual plans are best paired with fewer included minutes or a usage cap.
How should I price overage minutes?
Start from cost: price per extra minute = cost per minute ÷ (1 − target margin). At $0.158 a minute and a 60% target, that is $0.395. Many plans price overage just above the included-minute rate, which leaves heavy users at break-even or below.
What billed time do voice AI teams forget to count?
Hold and silence time are billed as call time. Retell bills calls under 10 seconds with a dynamic greeting as 10 seconds, and prompts over 4,000 tokens as proportionally more time (4,800 tokens is billed 20% more). Knowledge base, noise reduction and PII removal add $0.005–0.01 a minute each.
Sources and assumptions
- Component prices: Retell AI pricing, checked 22 November 2026: voice infrastructure $0.055/min, US telephony $0.015/min, voices and models as listed above.
- The example plan ($299 for 2,000 minutes, $0.20 overage) is illustrative, not a specific company's price.
- Not included: negotiated discounts, prompt caching, payment fees, support and hosting overheads. Margins here are gross margin on vendor costs only.
Related: Voice AI unit economics: real cost per minute · Voice AI pricing models · AI unit economics calculator