Voice AI cost per minute: what each platform lists
The per-minute price each voice AI platform publishes, what that price covers, and what you pay on top of it. A headline rate on one row isn't comparable with the next: some cover the whole call, some only the platform, so read the "on top" column first.
List prices only. Every number is LIST: what the vendor's own pricing page said when we checked it on 24 October 2026. Not a contract price, not a measured bill, and nothing here says anything about quality or latency.
What goes into one voice minute
A cascaded voice agent pays for five things each minute: the platform that runs the call, speech-to-text, the LLM, text-to-speech, and the phone line. Some vendors bundle several into one rate; others charge the platform fee and pass the rest through.
- Bundled or not. Bland, Deepgram's Voice Agent API, Mistral's voice agent API, Bolna and Sarvam include the models in the rate. Retell, Vapi and LiveKit add the LLM, and often speech, per model.
- The phone line. Almost no rate here includes telephony. On your own carrier or SIP trunk, the carrier bills you; on the vendor's numbers, a per-minute or per-pulse charge is added.
- Rounding. Retell bills to the second, and OpenAI's GPT-Live per second. Sarvam rounds each call up to whole 60-second pulses, and Bolna's Pilot plan bills in 30-second pulses, so short calls cost more than the rate suggests.
- Plan fees and minimums. Some rates need a monthly plan fee or a prepaid minimum, listed in the last column.
Per-minute list prices, in US dollars
Every rate and fee in this table is LIST, as published on the vendor's page on 24 October 2026.
| Vendor | Listed rate | What the rate covers | On top of it | Plan fees, minimums, billing |
|---|---|---|---|---|
| Retell AI | $0.055/min | Retell's voice infrastructure: the conversation engine that runs the call | LLM $0.0016–$0.64/min by model (GPT-5.6 Luna $0.0064, GPT-5.6 Terra $0.064). Text-to-speech $0.015/min for most voices, $0.04–$0.10 for ElevenLabs voices. Telephony on Retell's numbers by country; none from Retell on your own SIP trunk. Add-ons $0.005–$0.01/min each. Retell quotes $0.07–$0.31/min for a voice agent, depending on the LLM and voice; that range isn't an all-in rate with telephony. | No minimum. Billed to the second. 20 concurrent calls included, then $8 per concurrent call a month. Numbers $2 a month. |
| Vapi | $0.05/min | Vapi hosting and orchestration, prepaid | Speech-to-text, LLM and text-to-speech at provider cost; Vapi's examples are Deepgram $0.0095–$0.0099, OpenAI $0.0077–$0.0452 and ElevenLabs $0.0146–$0.0238 a minute. Telephony billed by the carrier, for example Twilio $0.008 inbound and $0.014 outbound a minute; no Vapi charge on Vapi telephony or SIP. | Usage only needs no plan. Core $29 a month. Pro is 10% of the hosting fee with a $999 a month minimum. |
| ElevenLabs Agents | $0.08/min beyond plan minutes ($0.16/min burst) | Hosting, speech-to-text, text-to-speech, knowledge base and RAG | LLM billed by model and usage. Telephony provider at cost. Text messages $0.003 each. | Plans from $6 to $990 a month include 75 to 12,375 minutes; the entry plan includes 15. |
| Deepgram Voice Agent API | $0.075/min Standard | Speech-to-text, LLM, text-to-speech and orchestration | Telephony isn't listed. Bring your own text-to-speech or LLM: $0.065/min; both: $0.050/min. Advanced tier $0.163/min. | Billed on websocket connection time. Pay-as-you-go rates, with no minimum. The Growth plan, which needs a commitment from $4,000 a year, lists lower ones ($0.068 Standard, $0.051 bring-your-own text-to-speech, $0.059 bring-your-own LLM, $0.041 bring both). |
| Mistral voice agent API (Voxtral 3) | $0.04/min with Mistral Small 4; $0.08/min with a larger model | Speech-to-text, the reasoning model and text-to-speech in one pipeline | Telephony isn't listed. | Announced 22 October 2026. No minimum stated. |
| OpenAI GPT-Live 1 | $0.05/min | A live voice session | "Backend model and tool usage is charged separately." Telephony isn't listed. | Billed per second, not rounded up to a whole minute. |
| OpenAI gpt-realtime-2.1 | No per-minute price | A speech-to-speech model | Priced per token: audio $32 in and $64 out per million tokens; the mini model $10 and $20. Telephony isn't listed. | Your cost per minute depends on how much each side talks. |
| Bland | $0.14/min Start; $0.12/min Build | LLM, speech-to-text and text-to-speech | Telephony on your own carrier, or Bland's at pass-through cost. Transferred minutes $0.05 (Start) or $0.04 (Build). | Start has no platform fee and a cap of 100 calls a day. Build adds a $299 monthly platform fee. |
| Synthflow | No public per-minute price | – | – | Enterprise contracts from $30,000 a year, scoped on volume, concurrency, telephony and integrations. |
| LiveKit Cloud | $0.01/min agent session, beyond plan minutes | Running the agent session | Speech-to-text, LLM and text-to-speech per model (for example GPT-5.6 Luna $0.0008/min, GPT-5.6 Sol $0.0155/min). A LiveKit US local number $0.01/min. | Plans include 1,000 to 50,000 agent minutes a month. |
| Bolna (India) | $0.06/min pay as you go (about ₹5.52) | Speech-to-text, LLM and text-to-speech, with Bolna's bundled models | "Telephony and the platform fee are added per minute"; the amounts aren't public. | About $0.045/min with larger top-ups. Pilot: $500 for 12,000 minutes or $300 for 6,500, billed in 30-second pulses. |
Per-minute list prices, in rupees
Every rate and fee here is LIST, in rupees as Sarvam publishes it.
| Vendor | Listed rate | What the rate covers | On top of it | Plan fees, minimums, billing |
|---|---|---|---|---|
| Sarvam | ₹3.50/min pay as you go; ₹3.25 Pro, ₹3.15 Growth, ₹3.00 Business (excluding GST) | The agent and the speech, language and voice models behind it | Calls on a Sarvam number ₹0.20 per 30-second pulse; numbers ₹59 per 30 days. Bring your own telephony on Business and Enterprise: no Sarvam charge, your provider bills you. | 60-second pulses, each call rounded up. Paid plans have a monthly minimum that comes back as credits. |
From a list price to your cost per minute
A list price is where the arithmetic starts. Your all-in cost per connected minute also depends on call length, silence, how long your prompts are, retries, telephony rounding and any contract discount, and only your invoices and call logs show those.
The worked example prices one voice stack three ways from list prices, with every number tagged. The Voice Cost Audit does the same on your own invoices and calls.
Prices change often. Each vendor name links to the page we read on 24 October 2026; check it before you plan. Spotted a price that has changed? Email poojith@illate.dev and we'll update the row.