How much does an AI voice agent cost per minute?

A complete AI voice agent costs about 8 to 13 cents per minute in the published vendor examples we checked ¹, before a phone carrier charges for the line. Add a US outbound phone line at Twilio's rate and the example comes to about 12 cents per minute ². Those are examples, not quotes. The real cost depends on the models you choose, how long each call runs, and what the agent does with tools.

Published examples price a complete agent per minute

Vapi's pricing page gives an example for 1,000 minutes a month. It uses Deepgram transcription, OpenAI and ElevenLabs models, and the total is $82 to $129 ³. Divided by 1,000 minutes, that is 8.2 to 12.9 cents per minute ¹.

Retell's pricing calculator gives a different default setup. Its example totals $0.11 per minute, made up of a language model cost, voice infrastructure and text to speech ⁴. Both examples set telephony to zero, so neither includes the carrier's charge for the call itself.

These examples come from vendors and may leave costs out, so treat them as a floor for a real estimate. They show the order of magnitude: about 8 to 13 cents for the agent ¹, plus the line.

A real phone line adds a carrier charge

When an agent places or receives a real call, a carrier bills for the line. Twilio's US pricing page lists $0.0140 per minute for calls made to the United States and Canada, with pricing current as of August 2026 ⁵. Adding that rate to Retell's example gives about 12.4 cents per minute ².

That sum combines two vendors' figures, so use it as a rough check. The page lists calls by direction, so check the rate for the direction your agent uses.

The platform fee is only one line

Vapi charges $0.05 per minute for hosting on its pre-paid plan, and its speech and language models are billed at provider cost ⁶. So the hosting fee and the model bills arrive as separate lines, and the model line depends on the model you choose.

A voice session can also be priced on its own. OpenAI lists its gpt-live-1 voice sessions at $0.05 per minute ⁷. Ask whether the backend model and tool usage are billed on top of that rate, because the session price is one piece of a call, not the whole call. The design guide explains why the two designs bill differently.

It depends on call length, model and tools

• Call length. Each minute bills, so a 10-minute call costs ten times a one-minute call at the same rate. Long calls, such as scheduling with several questions, change the total more than any single rate.
• Model choice. A larger language model can cost more per minute than a smaller one. A cheaper model may fail more calls, so check the cost of a failed call too.
• Tools. A booking lookup, a database query or a payment step adds its own charge and adds time to the call.
• Volume. The Vapi example is a monthly total for 1,000 minutes. Ask each vendor for the price at your expected volume, because the examples may not match it.

Speech recognition also has a price. For batch transcription of recorded audio, AWS lists $0.006 per minute in its worked example ⁸. A live agent uses a streaming recognizer, which may be priced differently, so check the streaming rate before you add it up.

How to estimate your own cost

1. List every line: hosting or platform fee, language model, recognizer, voice, phone line and tools.
2. Price a realistic call using your average call length and the number of calls per month.
3. Check the unit of each price. Some are per minute, some per second, some per token and some per month.
4. Add a margin for failed calls, retries, transfers and calls that run long.
5. Recheck after any vendor change, because the prices on these pages change.

For the dictation side of the same arithmetic, see our guide to AI dictation cost. For what a voice agent is, see the voice agent glossary entry, and for the main language model line, the large language model entry. The guide to AI voice agents, how they work and where they fail covers the whole system, and the voice and AI overview covers the subject.