What the AI voice platforms actually publish, per minute

There is a companion to this page. What AI voice agents actually cost to run explains the cost model, the four metered components and which of them scale with call volume, and it deliberately quotes no figures because figures date.

This is the other half. The figures, dated, with a link on every row.

Every number below was read off the vendor’s own pricing page on 16 September 2026. Where a total is a sum of published components, the arithmetic is shown so you can check it rather than trust it.

The headline rates, and why they are not comparable

Platform	Published headline	What that number covers
Vapi	$0.05 / min	Vapi hosting only. Speech, model and telephony pass through
Retell	$0.07 to $0.31 / min	An all-in range across model choices
ElevenLabs Agents	$0.080 / min	Speech in and out. Language model billed separately

Three numbers, three different scopes. Putting $0.05, $0.07 and $0.080 in a row next to each other and picking the smallest is the single most common mistake in this comparison, and it picks the one with the most left to add.

Vapi, component by component

Vapi publishes its own fee and the pass-through cost of each provider separately.

Component	Published rate
Vapi hosting, pre-paid	$0.05 / min
Transcription, Deepgram	$0.0095 to $0.0099 / min
Model, OpenAI	$0.0077 to $0.0452 / min
Voice, ElevenLabs	$0.0146 to $0.0238 / min
Transport, Telnyx	$0.0055 / min
Transport, Twilio outbound	$0.014 / min

Summing the published components gives the range an actual call lands in. At the bottom of every range, with Telnyx transport, $0.05 + $0.0095 + $0.0077 + $0.0146 + $0.0055 = $0.0873 per minute. At the top of every range, with Twilio outbound, $0.05 + $0.0099 + $0.0452 + $0.0238 + $0.014 = $0.1429 per minute.

So the platform advertised at $0.05 runs between roughly $0.087 and $0.143 a minute depending on which model and carrier you pick. The model is the component with the widest spread, nearly six times between its published floor and ceiling.

Retell, and what its floor actually is

Retell publishes an all-in range and the components behind it.

Component	Published rate
Retell voice infrastructure	$0.055 / min
Text to speech, Retell platform voices	$0.015 / min
Model, GPT 4.1	$0.045 / min
Model, Claude 5 Sonnet	$0.08 / min
Telephony, US Twilio example	$0.015 / min

The published floor of $0.07 is infrastructure plus speech and nothing else. $0.055 + $0.015 = $0.07 exactly. That is the price before the language model and before telephony, which is to say before the two components every real phone call needs.

A configuration that actually answers a call adds both. With GPT 4.1 and the Twilio example, $0.055 + $0.015 + $0.045 + $0.015 = $0.13 per minute. With Claude 5 Sonnet in place of GPT 4.1, $0.165 per minute.

ElevenLabs Agents, where the tiers buy concurrency rather than a discount

ElevenLabs publishes one per-minute rate and a set of monthly plans.

Plan	Monthly	Included minutes	Concurrent calls
Free	$0	15	4
Starter	$6	75	6
Creator	$22	275	10
Pro	$99	1,238	20
Scale	$299	3,738	30
Business	$990	12,375	40

Included and additional minutes are both $0.080. Exceeding your concurrency allowance is billed at a burst rate of $0.160 per minute, double. Text messages are $0.003 each.

Every tier prices its included minutes at exactly $0.080. 75 x $0.080 = $6. 275 x $0.080 = $22. 1,238 x $0.080 = $99.04. 3,738 x $0.080 = $299.04. 12,375 x $0.080 = $990. The plans are prepaid minutes at the list rate, so moving up a tier buys concurrency and buys nothing off the per-minute price.

That matters more than a discount would for an inbound use case. Concurrency, not minutes, is what fails on the morning everyone calls at once.

Concurrency, the cost nobody budgets for

Platform	Included concurrency	Cost to add more
Vapi, no success package	4	$10 / line / month
Vapi, Core	10	$10 / line / month
Vapi, Pro	30	$10 / line / month
Retell, pay as you go	20	$8.00 / concurrency / month
ElevenLabs, by plan	4 to 40	Burst billed at $0.160 / min

Three vendors, three models for the same constraint. Vapi and Retell sell it as a monthly line item. ElevenLabs does not sell it at all, it just charges double per minute once you exceed it, which is the only one of the three where a traffic spike degrades into cost rather than into rejected calls.

The telephony floor underneath all of it

Every platform above passes telephony through, so it is worth knowing what the carrier charges. From Twilio’s US voice pricing, read 16 September 2026.

Item	Published rate
Outbound, US and Canada local and toll-free	$0.0140 / min
Inbound, to a local number	$0.0085 / min
US local phone number	$1.15 / month

Inbound is meaningfully cheaper than outbound per minute, which is a quiet argument for the receptionist use case over the outbound dialler one on cost alone.

Free credits, for evaluating before committing

Vapi publishes $5 in free credits. Retell publishes $10 with no commitment. ElevenLabs offers 15 minutes a month on a free tier at 4 concurrent calls.

At roughly $0.13 a minute all in, $10 of credit is about 75 minutes of real testing, and $5 is about half that. Enough to validate a conversation design, not enough to load test one.

What we could not verify, and did not quote

• ElevenLabs language model and telephony costs. The pricing page states these are billed separately, “based on usage and varies by model”, and gives no figure. So the $0.080 cannot be turned into an all-in number from that page alone.
• A contradiction on that same page, left as we found it. The feature list shows telephony included on all plans, while the billing note says telephony provider costs are billed separately. We are not resolving that by picking the reading we prefer. Ask them before budgeting on it.
• GoHighLevel’s own AI voice pricing is not in this comparison. It is billed per enabled sub-account rather than per minute, which makes it a different shape of number entirely. It is covered, sourced the same way, in GoHighLevel cost and compliance figures.

Using these figures

Read the date. These were read on 16 September 2026, every one of these vendors has changed pricing inside a quarter before, and a per-minute rate quoted without a date is worth nothing. Follow the links before you put a number in a proposal.

And where a total here is a sum, the arithmetic is written out on purpose. Check it.