Voice AI pricing, by the minute
A live voice agent is billed for the minutes it is on the call. Pick a voice model, set how long your calls run, and read the month off the slip.
from$0.05a minute, 5 credits, on gpt-live-1
Work out your own call
Worked example: a 3-minute call on nova-sonic-2 is 3 × $0.064 = $0.192, which is 19.2 credits or ₹19.2. 1,000 of those calls in a month spend 19,200 credits.
Voice model cost
- $0.064 × 3 min
- $0.192
- One call
- 19.2 credits₹19.2
- × 1,000 calls
- $192
- A month of calls
- 19,200credits
- ₹19,200 or $192
Phone network fees are not in this total. They are billed by Meta or your carrier.
What the minute pays for
One per-minute rate covers the model listening, thinking and speaking for that minute. It is spent from the same credit balance as everything else, at 1 credit = ₹1 = $0.01. CallMissed adds no separate orchestration fee.
| Voice model | Per minute | Credits |
|---|---|---|
| gpt-live-1Speech-to-speech, billed per second | $0.05 | 5 |
| nova-sonic-2Speech-to-speech | $0.064 | 6.4 |
| nova-sonicSpeech-to-speech | $0.071 | 7.1 |
| gpt-realtime-2.1-miniRealtime, smaller model | $0.117 | 11.7 |
| gpt-realtime-miniRealtime, smaller model | $0.118 | 11.8 |
| gpt-realtime-2.1Realtime, largest model | $0.375 | 37.5 |
What it leaves out
The phone network. On WhatsApp Business Calling, Meta states that all user-initiated calls are free, and that business-initiated calls are billed in six-second pulses by the country being called. A SIP or PSTN number is billed by your carrier.
Read Meta’s calling rate rules
Plans, monthly credits, and the token, audio-hour and character rates for a cascaded speech-to-text, LLM and text-to-speech agent are on the main pricing page.
Or one flat rate for the whole agent
A voice agent plan covers listening, the language model and the voice as one rate a minute, with a 30-second minimum per call. A call that never connects costs nothing. The phone network is still the one separate line.
- Standard₹4/ minNatural Indian-language voices at the lowest price.
- Expressive₹5/ minRicher, more expressive voices across Indian and world languages.
- Best latency₹6/ minThe fastest replies and our best-sounding voices.
What other platforms publish
Read from each vendor’s own pricing page on 19 September 2026. The rates are not like for like: check what each one covers, and confirm with the vendor before you budget.
- Vapi$0.05 per minute hostingPlatform only. Model, transcriber, voice and telephony are passed through at provider rates.Source: Vapi
- Retell AI$0.07 to $0.31 per minuteRange depends on the voice, model and telephony chosen.Source: Retell AI
- Bland AI$0.14 per minute on Start, $0.12 on Build ($299 a month)The page says the rate covers LLM, speech-to-text and text-to-speech. Transfers are billed per transfer minute.Source: Bland AI
- ElevenLabs Agents$0.08 per minuteThe page says the LLM and any telephony are billed separately on top.Source: ElevenLabs Agents
- Ultravox$0.05 per minute, pay as you go30 minutes of free calls. SIP is listed as a separate per-minute line.Source: Ultravox
- Pipecat Cloud$0.01 per active minute (agent-1x)Agent hosting only. Models and telephony are billed separately.Source: Pipecat Cloud
- LiveKit Cloud$0.01 per agent session minuteSession minutes only. Speech-to-text, LLM and text-to-speech inference are billed separately.Source: LiveKit Cloud
- SynthflowNo per-minute rate publishedThe page lists enterprise contracts starting at $30,000 a year.Source: Synthflow
Questions about voice pricing
Is there a flat per-minute price for a whole voice agent?
Yes. A voice agent plan bills one rate a minute for listening, the language model and the voice: ₹4 Standard, ₹5 Expressive or ₹6 Best latency, with a 30-second minimum per call. A call that never connects costs nothing, and the phone network is billed separately.
What does the per-minute rate cover?
The voice model minute: listening, reasoning and speaking for that minute of the call, billed from your credit balance. CallMissed adds no separate orchestration fee on top. The phone network is the one separate line.
Is telephony included?
No. Voice model minutes and the phone network are billed separately. On WhatsApp Business Calling, Meta states that calls a customer starts are free, and calls your business starts are billed by Meta in six-second pulses by destination country. SIP and PSTN numbers are billed by your carrier.
Can a cascaded stack cost less than a speech-to-speech model?
It can. A cascaded agent pays a speech-to-text rate, an LLM token rate and a text-to-speech rate instead of one per-minute rate, and several of those models are on the free tier. The rates for each are on the main pricing page.
Why do published per-minute rates differ so much between platforms?
Because they cover different things. Some rates are hosting only, with models and telephony passed through. Others bundle the model and voice. Read what each rate covers before you compare the numbers, and confirm current pricing with each vendor.
How current are the other platforms’ figures?
Each one is what that vendor’s own pricing page showed on 19 September 2026, with a link to the page. Vendors change pricing often, so treat them as a starting point, not a quote.