guide
Voice AI Pricing in 2026: Two of Nine Publish a Rate
Every voice platform here can change something in your systems. Two of them will tell you what that costs.
Voice is the strangest corner of this category. Every platform in it can act on a system of record — look up an account, take a payment, change a booking — which is the most consequential capability anything here claims.
It is also the corner where almost nobody will tell you the price. Two of the nine publish a rate. The rest route you to a call.
| Category | Tools | Can change something | Publishes a price |
|---|---|---|---|
| Voice & Phone AI | 9 | 9 of 9 (100%) | 2 of 9 (22%) |
| Enterprise & CCaaS | 15 | 14 of 15 (93%) | 7 of 15 (47%) |
| Agent Assist & Copilots | 5 | 4 of 5 (80%) | 1 of 5 (20%) |
| AI Agents & Chatbots | 28 | 22 of 28 (79%) | 19 of 28 (68%) |
| E-commerce Support | 6 | 4 of 6 (67%) | 3 of 6 (50%) |
| Help Desk & Ticketing | 14 | 7 of 14 (50%) | 10 of 14 (71%) |
| QA & Conversation Analytics | 6 | 1 of 6 (17%) | 1 of 6 (17%) |
Ordered by the share that can change something, not by size.
Why per-minute is harder than it looks
Every other billing model in this directory meters an outcome or a seat. Voice meters time, and time is the one unit where the vendor controls the numerator. A slower model, a longer greeting, a caller left on hold inside the agent's flow — all of it bills.
So a per-minute quote is only comparable if you also know what counts as a minute, what happens during silence, and what happens when more calls arrive than your plan allows. Those three answers live in documentation, not on pricing pages.
ElevenLabs: $0.08, and the arithmetic is checkable
ElevenLabs publishes seven tiers on bundled call minutes. Divide any paid one by its included minutes and you get the same number to four decimal places: $6 for 75, $22 for 275, $99 for 1,238, $299 for 3,738, $990 for 12,375. All of them come to $0.0800 a minute.
The odd allowances are the tell — 1,238 and 3,738 were not chosen as round numbers, they were back-computed from a price and a rate. Which means the tiers are volume commitments rather than volume discounts, and the only place the page mentions a discount is Enterprise.
Two more meters sit underneath. LLM inference bills separately, by input and output tokens, varying with the model you pick. And burst pricing lets an agent exceed its concurrency cap by up to three times, with those calls charged at double. Neither appears in the tier table.
One clause goes the buyer's way and it is a real one: silent periods bill at 5% of the usual rate. Calls are full of silence, most per-minute pricing charges for all of it, and this does not.
Dialpad: seats you can see, and an agent you cannot price
Dialpad publishes seat pricing — $80, $115 and $150 per user per month, identical on both billing toggles, which is unusual in itself. Those figures sit behind a product dropdown that opens on the AI Agent panel, where no figure appears at all, which is why an earlier read of ours recorded the page as showing no prices.
The AI Agent itself is a pool of conversation credits quoted by sales, and it requires Support Advanced or Premium underneath. So the real floor for the agentic product is $115 a seat plus an unquantifiable amount.
What Dialpad does publish is the meter, and it is the clearest definition in the category: a conversation is billable only when the AI retrieves information or executes a real action. An agent that says hello and hands over is not a charge. That is a better-designed unit than most of the per-resolution models in this directory, and it is on a documentation page rather than a pricing page.
The other seven
PolyAI, Parloa, Replicant, Hyro, Interface.ai, Posh and Quiq publish no rate. Several have no pricing page at all — the URL 404s, which is a stronger statement than a page that exists and withholds.
The per-seat and per-minute figures circulating for these platforms in comparison posts are third-party. We do not carry them, because a number nobody can check against a vendor page is not a price, it is a rumour with a decimal point.
What to ask when there is no page to read
Three questions get you a comparable number faster than a discovery call. What is a billable minute — does it start at connect or at first word, and is silence charged at full rate? What happens when concurrent calls exceed the plan: queue, drop, or bill at a premium? And is model inference inside the rate or beside it?
A vendor that answers all three has thought about its unit. A vendor that cannot answer the first one has not finished designing the product, which is the same conclusion the transparency data points at across the whole directory: what predicts a published price is how settled the billing model is.
Rates read on each vendor's own pages in August 2026 and dated on the tool pages. Where a vendor publishes nothing we say Not published rather than reprinting a third-party estimate.
Frequently Asked
How much does an AI voice agent cost per minute?
Of the nine voice platforms in this directory, two publish a rate. ElevenLabs works out at exactly $0.08 a minute on every paid tier. Dialpad publishes seats at $80, $115 and $150 a user but quotes its AI Agent credits through sales. The other seven publish nothing, and the per-minute figures circulating for them are third-party rather than vendor-published.
Is a per-minute rate the whole bill?
Rarely. ElevenLabs meters LLM inference separately by token, varying with the model, and charges double for calls that burst above your concurrency limit. Dialpad's AI Agent needs a $115 seat underneath it. Ask specifically whether model inference is inside the rate or beside it, because it is the meter most likely to be missing from the page.
Why is voice AI pricing so hidden?
It is the least settled billing model in the category. Seats have been priced the same way for twenty years and list easily; nobody has agreed what a voice minute should include. Across the whole directory the strongest predictor of a published price is not company size or who they sell to, it is whether the vendor has decided what its unit is worth.
What counts as a billable minute?
It varies and it is worth pinning down. ElevenLabs bills silence at 5% of the normal rate, which is a genuine concession most vendors do not make. Dialpad publishes the sharpest definition anywhere here: a conversation is billable only when the AI retrieves information or executes a real action, so an agent that greets and hands over is free.
Can I try a voice agent without talking to sales?
One. ElevenLabs has a free tier with 15 minutes of calls and 4 concurrent calls. It is the only customer-facing voice agent in this directory with a self-serve path, which in a category of nine is the finding rather than a footnote.
Tools Mentioned
Full reviews, pricing tiers and where each one breaks.
ElevenLabs Agents
Voice-first agents from the speech company, sold on bundled call minutes with a free tier and no sales call. Acting on your systems is a webhook you write.
Dialpad Ai
Business phone system with an autonomous AI agent attached. The seats are published; the agent is quoted.
PolyAI
Enterprise phone agents for brands with very high call volumes. No self-serve path.
Parloa
Contact-centre phone automation focused on sounding human rather than on channel breadth.
You Can Also Look Into
Only 38 of 83 AI Customer Service Vendors Publish Their Pricing
The full count across 83 platforms, the categories where transparency collapses entirely, and what a hidden price actually costs you.
ElevenLabs Agents Pricing: Every Tier Is $0.08 a Minute
Divide any paid ElevenAgents tier by its included minutes and you get $0.0800. Starter, Creator, Pro, Scale, Business — identical. The tiers are volume commitments rather than volume discounts. Here is the arithmetic, the two meters the pricing page does not show, and the one clause that goes in your favour.
SOURCES
WRITTEN BY AR · UPDATED 2026-08-17
I read the fine print. Vendor pricing pages, billing definitions, terms, funding filings and acquisition notices — then I do the arithmetic nobody publishes: what a platform actually costs at your volume, what its headline metric is really counting, and who owns it now. I do not run benchmarks, and no page here pretends otherwise.