What a minute of an AI phone call costs across Vapi, Retell and ElevenLabs

A headline per-minute price tells you little about what an AI phone call costs. The figures in the StackVersus voice directory, checked on 2026-10-30, show three different billing models. Vapi and Retell charge a platform fee and pass through the model components. ElevenLabs Agents bundles voice and transcription into a per-minute price and bills the language model on top. Our reading is that the lowest-looking fee is rarely the lowest total cost. Volume, call length and the voice you need will decide the answer.

Vapi’s low fee is one line of several

Vapi charges a platform fee of $0.05 per minute, prorated to the second. Its pricing copy says transcriber, model, voice and telephony costs are “charged at-cost” on top of that fee. Telephony, SIP, WebSockets and Daily WebRTC are listed as free.

Vapi’s own calculator defaults show how the stack adds up. Deepgram speech-to-text runs $0.0095 to $0.0099 per minute, OpenAI’s LLM runs $0.0077 to $0.0452 per minute, and ElevenLabs text-to-speech runs $0.0146 to $0.0238 per minute. Its 1,000-minute example totals $82 to $129 per month.

This suggests the $0.05 headline is a floor, not a quote. The final figure depends on the model, voice and transcriber you select, and each one moves the bill.

Vapi’s plan structure also matters for volume. The free No Success Package allows 4 concurrent calls. Core, at $29 per month, allows 10. The Pro plan starts at $999 per month for 30 concurrent calls. Teams that need concurrency will likely cross into a fixed monthly commitment well before they reach a large call volume.

Retell’s range is wide by design

Retell bills per minute, timed to the nearest second and summed across the billing cycle, with no per-call rounding up. Its Voice Infra platform fee is $0.055 per minute. The advertised all-in range is $0.07 to $0.31 per minute.

That gap is the most useful number in this comparison. The bottom of the range likely reflects inexpensive voices and models. The top likely reflects premium voices and heavier prompts. Voice choice drives most of the spread. The directory lists voice options from $0.015 per minute up to $0.10 per minute, so a reader who needs a specific premium voice should expect a higher bill.

Retell’s billing also has gotchas that can raise cost. Calls under 10 seconds that use a dynamic opening message are billed a 10-second minimum. Prompts over 4,000 tokens scale billed duration, using the prompt’s LLM token count divided by 4,000. Silence and hold time are billed because speech-to-text keeps listening. Short-call and long-prompt workloads should be modeled before a team commits.

ElevenLabs trades itemization for predictability

ElevenLabs Agents includes call minutes in each plan and charges $0.080 per minute for additional calls on all self-serve plans. Burst pricing above the concurrency limit is $0.160 per minute. Speech-to-text is listed under included features, and telephony passes through at cost.

The plan math is consistent. Creator at $22 for 275 minutes works out to $0.08 per included minute. Pro at $99 for 1,238 minutes and Business at $990 for 12,375 minutes both come to $0.08 per minute. The paid tiers price minutes at the same rate that overage is billed, so a team that knows its volume can forecast its voice cost precisely.

The LLM is the exception. ElevenLabs passes through third-party LLM costs “at the provider’s published rate, with no markup,” but select Gemini and Claude models carry a 10 percent uplift on Vertex AI regional rates. For example, gpt-4o-mini lists at $0.15 per million input tokens and $0.60 per million output tokens. Teams that want a particular model should confirm its current rate on the vendor’s site.

Vendor	Billing model	Voice and transcription	LLM
Vapi	$0.05/min platform fee plus at-cost components	Itemized	Itemized
Retell	Usage-based per minute, advertised $0.07 to $0.31 all-in	Itemized	Itemized
ElevenLabs Agents	Included minutes, $0.080/min overage	Bundled	Passed through at provider rate

What to do

Start with your expected monthly call minutes, average call length and concurrency needs. Model the LLM and voice you plan to use on each vendor’s calculator. Check the minimum billable duration and prompt size, because both can change the bill for short or prompt-heavy calls. Confirm current rates on each vendor’s pricing page before you decide, since figures can change.

---