AI Voice Agent Cost per Minute, All-In: Vapi vs Retell vs Bland (LLM, Voice and Telephony Split)

Every voice-agent platform puts one number on its pricing page. Vapi says $0.05 a minute. Retell says from $0.07. Bland says $0.14. They aren't the same thing. One is a hosting fee, one is a starting point for a range that runs to $0.31, and one is a bundle that covers the AI but not the call itself.

So we read each vendor's pricing page and docs on 2 October 2026 and split one minute of a phone call into its five parts: the platform fee, speech-to-text, the LLM, text-to-speech and the phone line. We haven't built or run a voice agent on any of these platforms, and none of them pay us. The chat side of the same question, cost per resolved support conversation, is in our AI customer service cost guide.

What one minute costs, all-in

A minute of a US inbound call costs about $0.08 on the raw APIs, $0.11 on Vapi, Retell or ElevenLabs Agents, and $0.15 on Bland's Start plan, with a mid-size LLM. The gap between the cheapest and dearest route is less than 7 cents a minute.

How we filled each column:

• LLM: $0.024 a minute, which is what Retell's pricing page lists for GPT-5.4 mini. We used the same figure on every route that bills the model separately, so the LLM line is equal. Vapi's own calculator puts OpenAI models at $0.0077 to $0.0452 a minute, which brackets it.
• Vapi speech: Deepgram transcription at $0.0099 and ElevenLabs voice at $0.0146, from Vapi's calculator. Vapi's docs say it transcribes both the caller and the assistant, so the transcription line covers two audio channels.
• Retell: its "Voice Infra" at $0.055 has no separate speech-to-text line, so transcription sits inside it. Its own platform voices are $0.015 a minute; ElevenLabs voices on Retell are $0.040.
• Phone line: Twilio's US rate for receiving a call on a local number, $0.0085 a minute. Retell sells its own US Twilio line at $0.015.

Two rows need a warning. The two cheapest, OpenAI GPT-Live and Deepgram's Voice Agent API, are raw APIs. You get the voice conversation, not a product: you write the code that answers the phone, connects tools, transfers calls and logs transcripts. Vapi, Retell, Bland and ElevenLabs sell that layer.

With a bigger model and a premium voice, the managed platforms move up. Vapi with a GPT-6 Sol-class model ($0.064 a minute) and ElevenLabs at the top of Vapi's range comes to about $0.16 a minute. Retell with GPT-6 Sol and ElevenLabs voices comes to $0.174.

What each vendor's headline price includes

Only Bland and Deepgram put the LLM inside their per-minute rate. Vapi's $0.05 covers hosting only. No vendor here includes the phone line.

In the vendors' own words:

• Vapi. The per-minute cost has "Four components: speech-to-text, the language model, text-to-speech, and transport. Each is billed at the provider's listed price with no markup. Vapi's $0.05/min hosting fee is added on top." Transport is "Charged by provider, not Vapi".
• Retell AI. Pay as you go is "$0.07-$0.31 / min for AI Voice Agents", with "No platform fees". The range is the sum of its component lines: Voice Infra, a voice, an LLM and telephony. Calls are "tracked to the nearest second", and after a transfer "the AI voice agent fee stops. Only the telephony fee continues".
• Bland. Its page lists "LLM No token charges · STT Real-time transcription · TTS Premium voices" as included, and adds: "Telephony is billed separately, on your own carrier or Bland's at pass-through cost." Bland doesn't publish its pass-through rate, so we used Twilio's.
• ElevenLabs Agents. "Additional call minutes cost $0.08 per minute… LLM usage is billed separately on top, based on the model you choose." On telephony: "that provider bills you directly for the carrier side of the call". Its monthly plans are prepaid minutes: Business is $990 for 12,375 minutes, which is $0.08 each.
• Deepgram Voice Agent API. Standard is $0.075 a minute, "calculated based on websocket connection time". Its docs put GPT-5.4 mini, Claude Haiku 4.5 and Gemini 3.5 Flash in the Standard tier. Larger models are Advanced, at $0.163.
• OpenAI GPT-Live. "$0.05" a minute, "billed per second", and "Backend model and tool usage is charged separately." OpenAI's cost guide adds that billable time includes moments "when the user speaks, the assistant speaks, both are silent, or the backend is working."
• Twilio ConversationRelay. $0.07 a minute on Twilio's US page. It handles the speech; you bring the model and pay Twilio's normal call minutes.
• Synthflow. No per-minute price: "Enterprise contracts start at $30,000 annually." That's $2,500 a month before a single call is priced.

Bland's own FAQ makes the same point about its rivals: it says Vapi's and Retell's advertised rates are "the platform fee only", and estimates production stacks at "roughly $0.13 to $0.30 per minute on Vapi and $0.11 to $0.25 per minute on Retell". That's a competitor's estimate. With a mid-size model and standard voices, our split lands at or just below the low end of it.

A 5-minute call, and the bill at 1,000 and 10,000 calls

A five-minute US call costs $0.41 to $0.74 depending on the route. At 1,000 calls a month that's $413 to $942; at 10,000 calls it's $4,125 to $6,724.

Bland's Start plan caps you at 100 calls a day, so 10,000 calls a month needs Build, which lowers the minute to $0.12 but adds $299 a month. Bland's docs also list a Scale plan at $499 a month and $0.11 a minute; the current pricing page shows only Start, Build and Enterprise.

What the per-minute math hides at volume is concurrency: how many calls can run at once. 10,000 five-minute calls is about 2,300 call-minutes per business day.

• Vapi usage-only allows 4 concurrent calls. Core, at $29 a month, raises it to 10; extra lines are $10 a month each.
• Retell includes 20 concurrent calls, then $8 a month per extra one.
• ElevenLabs runs 4 to 40 concurrent calls depending on the plan, and calls above your limit are billed at "burst pricing" of $0.16 a minute, double the rate.
• Bland allows 10 concurrent calls on Start and 50 on Build.

The LLM line is the one you control

The LLM is the only part of the minute that swings by more than 10 times. On Retell's price list it runs from $0.0016 a minute for GPT-5 nano to $0.32 for GPT-6 Astra. Speech and phone lines move by fractions of a cent.

A voice agent doesn't bill the model per minute directly. It sends the conversation to the model as tokens, several times a turn, and the platform converts that to a per-minute estimate. Vapi's docs spell out the assumptions behind its estimate: 5 model requests per turn, 150 output tokens a minute, and a 50% prompt-cache hit rate. They also warn that on "Long calls" the "Actual cost is often higher" because the "Growing conversation history is sent again with each request." If tokens are new to you, our explainer on what a token is covers why a longer prompt costs more on every turn.

Speech-to-speech models are priced differently again. OpenAI's gpt-realtime-2.1 bills audio tokens at $32 per million in and $64 per million out, where a user's audio is one token per 100 milliseconds and the assistant's one per 50. OpenAI's guide notes that "The entire conversation is sent to the model for each Response", so later turns cost more. Retell lists gpt-realtime-2.1 at $0.38 a minute, about 3.5 times our whole Retell stack with a text model.

Three ways to keep this line small, from the vendors' own guidance:

• Shorten the system prompt and tool definitions. Vapi calls this "the largest effect on model cost".
• Pick the smallest model that handles the call. A booking or routing call rarely needs a frontier model.
• Keep calls short. Every platform here bills by the second or minute, and Retell and OpenAI both bill silence and hold time.

The phone line: what Twilio charges in the US

The phone line is the smallest part of the minute but the one every headline leaves out. Twilio charges $0.0085 a minute to receive a call on a US local number and $0.014 a minute to make one.

Two details change the math. A toll-free inbound line costs more than double a local one, $0.022 against $0.0085, which adds about 7 cents to a five-minute call. And Vapi's free phone number is limited: its docs say free Vapi numbers are US-only and "inbound only", so an outbound campaign needs a number you import from Twilio, Vonage or Telnyx. Vapi lists Telnyx at $0.0055 a minute.

Extras on the Twilio side bill separately too: call recording at $0.0025 a minute, answering-machine detection at $0.0075 a call, and branded caller ID at $0.12 a call.

Costs the per-minute rate doesn't show

Most of the rest of the bill is fixed monthly fees and per-feature add-ons. They matter more at low volume.

• Compliance. Vapi's HIPAA add-on, a Data Processing Agreement plus a Business Associate Agreement, is $2,000 a month. ElevenLabs lists BAAs for HIPAA customers on Enterprise, Retell lists custom BAA terms on Enterprise, and Bland lists a BAA on Enterprise.
• Retell add-ons per minute. Knowledge base +$0.005, advanced denoising +$0.005, safety guardrails +$0.005, PII removal +$0.01. Four of them add $0.025 a minute, more than its LLM line.
• Transfers. Bland charges $0.05 a minute on Start ($0.04 on Build) once a call is handed to a human on its numbers. Retell stops the AI fee and keeps the telephony fee.
• Silence. Retell bills "the entire duration of the call because the speech-to-text engine remains active". OpenAI GPT-Live bills time when both sides are silent.
• Support and SLAs. Vapi's Pro package is 10% of hosting fees with a $999 monthly minimum, for a 99% uptime SLA and Slack support.

For comparison, Salesforce prices voice by action, not by minute. Its Agentforce page says "Agentforce Voice actions are 30 Flex Credits", which at $500 per 100,000 credits is $0.15 an action, with the Salesforce licenses and phone line on top. A five-minute call that runs four actions would be $0.60 before telephony. How that per-action and per-conversation billing plays out across support platforms is on our AI automation hub.

How to choose

If your voice agent is one piece of a wider agent build, the tools and frameworks for that are on our AI agents hub.

aliteq sells none of these products and earns nothing from these links. Every price here was read on the vendor's own page or docs on 2 October 2026.