What does a minute of voice AI really cost?
Bundled voice AI vendors quote one tidy per-minute price. Inside it are telephony at a markup, the AI model at a markup, and their margin. Enter your volume and see the same month priced on your own Twilio and AI engine accounts.
Your month
Defaults are illustrative. Edit anything to match your invoices.
Roughly 15, 60 or 300 calls a day, at three and a half minutes each.
Bundled vendor
Your own providers
WhautoVoice plan is picked automatically from all your minutes, AI and human together, because both are processed minutes: Starter US$39 (3,000 min), Growth US$179 (12,000 min), Scale US$549 (50,000 min). Above an allowance, extra minutes are added in blocks of 1,000 at US$5 (half a cent a minute). There is no overage.
Compared with a bundled vendor at the rates you entered.
| Carrier, AI-handled minutes | US$0 |
| Carrier, human minutes | US$0 |
| AI engine usage | US$0 |
| WhautoVoice Growth | US$0 |
| Extra minutes | US$0 |
| Total | US$0 |
Same Twilio. Same AI engine. No meter on top. Try it on your own numbers for 14 days.
Start free trialHow bundled voice AI pricing works
Most AI phone agent platforms sell a single per-minute rate. It is convenient, and it hides three separate costs: the carrier minute (what Twilio or a similar provider charges to connect the call), the AI usage (speech recognition, the language model and speech synthesis, or a single speech-to-speech model like OpenAI Realtime), and the platform's own margin. Because the vendor buys the first two wholesale and sells them retail, the published rate is usually two to four times the underlying cost. Many vendors also keep charging after the AI hands the call to a human, or sell a separate seat-based dialer for that.
What bring-your-own-key actually costs
With WhautoVoice you connect your own Telnyx, Twilio, Plivo or Vonage account and your own AI engine key (OpenAI, xAI Grok, ElevenLabs or Sarvam AI). The carrier bills you its published per-minute rate. The AI engine bills you its published usage rate. WhautoVoice charges a flat plan fee that includes an allowance of processed minutes, our fee for handling each call whether an AI agent or a person took it, with extra minutes at half a cent. The calculator adds those three lines together and compares the total with the bundled vendor's total at the same volume.
| Cost line | Bundled vendor | Bring your own keys |
|---|---|---|
| Carrier minute | Inside the per-minute rate, marked up | Your carrier's list price, billed by them |
| AI usage | Inside the per-minute rate, marked up | Your AI engine's list price, billed by them |
| Human agent minute | Metered at the vendor's AI rate after transfer, or not offered | Carrier minute plus a processed minute, half a cent above your plan |
| Platform | Margin inside the rate, sometimes a fee too | Flat plan from US$39 a month |
Methodology and assumptions
- Bundled total = AI-handled minutes x AI rate + human minutes x human rate + platform fee.
- WhautoVoice total = (AI-handled minutes + human minutes) x carrier rate + AI-handled minutes x AI engine rate + plan fee + any blocks of 1,000 extra minutes. There is no overage.
- Plan selection picks the cheapest combination of plan and blocks of 1,000 extra minutes (US$5 each, half a cent a minute) that covers all your minutes, AI and human together, because both are processed minutes. Above 50,000 minutes the Scale plan applies with blocks on top.
- Carrier defaults reflect typical US per-minute voice rates for programmable voice; international and toll-free rates are higher. Number rental (a few dollars a month per number) is excluded on both sides because it is the same either way.
- AI engine defaults approximate OpenAI Realtime API audio pricing at typical talk ratios. xAI Grok, ElevenLabs and Sarvam AI publish their own per-minute rates, so enter yours. The smaller realtime model is several times cheaper than the flagship. Rates change; check OpenAI's current price list.
- Taxes, SMS and recording storage are excluded on both sides.
Reading the result
At low volume the two totals are close, because a flat plan fee is a larger share of a small bill. The gap opens quickly with volume, and it opens faster if your team spends real time on the phone, because on WhautoVoice those minutes cost your carrier rate plus at most half a cent each, never an AI rate. If the calculator shows the bundled vendor ahead, that usually means very low AI volume and no human minutes. In that case the trial is still free, and the numbers will move as you grow.
Frequently asked questions
Published all-in rates for bundled AI phone agent platforms generally fall between US$0.07 and US$0.30 per minute once telephony, speech, language model and platform fees are added together. Many also meter minutes after a call is transferred to a human. Use the presets or type your quote.
Roughly the carrier's per-minute voice rate, about one to two US cents in the United States, plus OpenAI Realtime usage, typically a few cents a minute for the smaller realtime model and around six to ten cents for the flagship model, plus a flat platform fee. Rates change, so check current price lists and edit the defaults.
Because many bundled vendors keep the meter running after the AI transfers a call to a person, or charge separately for a human dialer. WhautoVoice counts those minutes as processed minutes, inside the plan allowance or half a cent each beyond it, never at an AI rate, so the gap widens as your team spends more time on the phone. Set the bundled human rate to zero if your vendor does not charge for it.
We built it, so read it with that in mind. Every input is editable and the formula is printed above. If your bundled quote is genuinely US$0.05 a minute all-in with free human minutes, type that in and the calculator will tell you so.