CanaricsWhat a voice AI agent actually costs in 2026
Four pricing models compete for the same buyer: developer platforms billed per minute, enterprise contracts, managed vendors with setup fees and metered minutes, and flat subscriptions. The headline numbers are not comparable until you add what each one leaves out. Named-vendor prices were checked on 6 September 2026 and link to each vendor's own pricing page; the unnamed local offers are described under the worked example.
The four pricing models
1. DIY platforms (per minute, you build the agent). Vapi, Retell, and Bland sell infrastructure to developers. The advertised rate is low, but on Vapi and Retell it covers only the platform layer: the language model, speech-to-text, voice synthesis, and telephony are billed on top, and the agent itself, its prompts, integrations, and monitoring, is your engineering work to build and maintain.
2. Enterprise contracts. Synthflow now starts at $30,000 per year with pricing scoped per deployment. Built for contact-center scale, not for a company with one phone line.
3. Managed vendors billed per minute. Common in local European markets: a setup fee, a monthly subscription, and a per-minute meter on top. The entry price looks small; the meter is where the money is.
4. Flat subscription (Canarics). One monthly price with a large minute bundle included, telephony and analytics included, no setup fee on the self-serve path. The trade-off is a higher entry price than a bare meter.
Global platforms, side by side
| Vendor | Advertised price | What it excludes | Realistic all-in | Who it is for |
|---|---|---|---|---|
| Vapi | $0.05/min platform fee | The AI itself: Vapi ships no model, you connect your own LLM, speech-to-text, and voice (passed through at provider cost, or billed to you directly on your own keys), plus telephony and your build and maintenance work; HIPAA is a $2,000/mo add-on | $0.11 to $0.20/min computed at published provider rates, before your engineering time (see the worked example); Bland's comparison estimates $0.13 to $0.30 | Developer teams building their own agent |
| Retell AI | $0.07 to $0.31/min | Telephony ($0.015/min), add-ons (knowledge base, guardrails, PII removal billed per minute each), build and maintenance work | $0.13/min computed from Retell's own component prices (voice infra $0.055, mid-tier LLM $0.045, platform voice $0.015, telephony $0.015), before your engineering time; premium voices and top LLMs push it to $0.19+ | Developer teams building their own agent |
| Bland AI | $0.14/min ($0 platform fee), or $0.12/min plus $299/mo, or $0.11/min plus $499/mo | Telephony (where passed through), transfer minutes ($0.05/min on Start), build and maintenance work; warm transfers and guardrails are Enterprise-only | $0.155/min on the Start tier with telephony, before your engineering time; the $299 tier only beats Start above roughly 15,000 min/mo | Developer teams; the most complete advertised rate of the three |
| Synthflow | Enterprise contracts from $30,000/yr | Final pricing scoped per call volume, concurrency, integrations, and launch support | From roughly $2,500/mo equivalent | Contact-center scale deployments |
| Canarics | €249/mo inbound or €349/mo outbound (inbound included), 1,000 minutes included | Nothing at typical volume: number, telephony, 40+ languages, transcripts, caller history, dashboard, and upgrades are in the price | €249 or €349 flat; in practice the subscription is the whole bill, since most of our clients stay under the included 1,000 minutes | Businesses that want a working agent, not a toolkit |
Vendor prices in USD as published on 6 September 2026. All-in rates for Vapi and Retell are computed from published component prices (the method is under the worked example below), since neither platform publishes an all-in number; Bland's own pricing page estimates the same stacks at $0.13 to $0.30 (Vapi) and $0.11 to $0.25 (Retell), consistent with our math. Canarics extra minutes are billed in 100-minute blocks at the same per-minute rate (€24.90 inbound, €34.90 outbound per block).
A worked example: 1,000 minutes per month
Roughly 330 answered calls at the 3-minute average we measure across clients, a normal month for a small company that stops missing calls.
The platform bill is the smallest of three numbers a DIY stack costs you. The other two are the build (a production voice agent takes weeks of engineering: telephony wiring, prompt design, interruption handling, testing, edge cases) and the running of it (models change, prompts drift, calls fail in new ways; someone has to watch it). Comparing a platform bill to a managed price is comparing an engine to a car.
| Option | Platform bill / subscription | Build (one-time) | Running it (people, monthly) | Year-one monthly total (setup and labor averaged in) |
|---|---|---|---|---|
| Vapi (computed stack) | $110 to $195: platform $50, plus the AI you connect yourself (speech-to-text about $10, LLM $20 to $80, voice $15 to $40) and telephony about $15, across four or five separate vendor accounts | About €8,000 of your engineering (16 days at €500/day) | About €500 (10 hours of tuning, testing, and monitoring at €50/hour) | About €1,300 |
| Retell AI (computed stack) | $130 to $190: voice infra $55, mid-tier LLM $45, platform voice $15, telephony $15 | About €8,000 of your engineering | About €500 | About €1,300 |
| Bland AI (Start tier) | About $155: $140 talk time plus telephony pass-through | About €8,000 of your engineering | About €500 | About €1,300 |
| Synthflow (Enterprise) | from about $2,500 equivalent | Included in contract | Included | From about €2,500 |
| Typical EU vendor, metered (€149/mo + €0.39/min) | €539 | €1,500 setup | Included | €664 |
| Typical EU vendor, high base (€500/mo + €0.03/min) | €530, where the product is a scripted campaign call rather than an open conversation | €400 setup | Included | €563 |
| Typical carrier voice bot (€250/mo + €0.05/min) | €300, a product category built for announcement playback rather than two-way conversation | €300 setup | Included | €325 |
| Canarics | €249 (inbound) or €349 (outbound, inbound included) | €0 (custom integrations, if any, €500 to €1,000 one time) | Included | €249 or €349 |
DIY platform bills are computed from the vendors' published component prices: Retell lists every component on its pricing page (voice infra $0.055/min, LLMs $0.003 to $0.16/min, voices $0.015 to $0.04/min, telephony $0.015/min); Vapi bills models at provider cost, so we use the same provider rates on top of its $0.05/min platform fee; Bland's rate includes the models, with telephony passed through. Labor assumptions are stated in the cells (€500/day, €50/hour, 16 build days spread over year one, 10 hours a month of operations) so you can substitute your own rates; a template agent can be live in days, but 16 days and 10 hours a month is the optimistic end of what a production agent with integrations, interruption handling, and monitoring took us, and we run this stack for a living (what running one actually involves). USD figures are treated at parity with EUR in the totals; converting at the current exchange rate would lower the platform stacks by roughly 10 to 15 percent in euro terms and changes no conclusion. The three unnamed rows are pricing structures currently offered in the Baltic market, presented as archetypes rather than named offers because those prices are not on stable public pages we can link and terms vary by customer.
The pattern to watch: where the money hides
Each model carries its cost in a different place. Per-minute platforms split it across the stack: the advertised rate is one of four or five meters running at once. Metered managed vendors put it behind the anchor: a small subscription in front, €0.39 a minute behind it, and a four-figure setup fee at the door. Enterprise contracts carry it in scope. A flat subscription carries it all in one number, which is why it looks more expensive at the first glance and cheaper on the third invoice.
Against the €149 plus €0.39/min structure, €249 flat with 1,000 minutes included breaks even at roughly 250 minutes per month, about 85 average calls. At 300 minutes the metered offer costs €266, at 1,000 minutes €539, plus the €1,500 setup in year one. A company that misses roughly a third of its inbound calls (34.7% measured at one live client, 30.3% fleet-wide) sits well above the break-even line.
Put in year-one monthly totals at 1,000 minutes: a self-built platform stack runs about €1,300, an enterprise contract from €2,500, the metered local offer €664, and a flat subscription €249. The conclusion is not knife-edged on the volume either: at 2,000 minutes Canarics is €498 (€249 plus ten overage blocks) and is still the lowest column. The cheapest advertised number and the cheapest total belong to different columns.
Where the other models win
Very low volume. Under roughly 250 minutes a month, a metered offer with a small subscription is genuinely cheaper. If your phone rings twice a day, a €249 flat plan is not for you yet.
You have voice AI engineers. If building conversational agents is your team's actual job, Vapi, Retell, or Bland at $0.11 to $0.30 per minute all-in gives you control no managed product matches. That is a build-vs-buy decision, not a price comparison.
Mass notification at scale. A high-base, €0.03-per-minute robocall platform undercuts a €249 flat subscription on paper above roughly 2,300 minutes a month. What the €0.03 minute buys is a scripted announcement, not a conversation that books, answers, and transfers; if a recorded message is all the job needs, it is the right tool.
Contact-center scale. At hundreds of thousands of minutes with custom concurrency, routing, and compliance review, an enterprise contract with scoped pricing is the honest structure, and per-unit prices below any list price are negotiable.
Canarics pricing in full
- Inbound receptionist: €249 per month per licence, 1,000 minutes included.
- Outbound: €349 per month per licence, 1,000 minutes included; covers outbound campaigns and follow-up calls, and inbound is included, so callers who ring back reach the same agent.
- Extra minutes: billed in 100-minute blocks at the same per-minute rate (€24.90 or €34.90 per block), so a busy month does not change the unit price.
- Included: phone number and telephony, conversations in more than 40 languages with mid-call switching (Latvian, Lithuanian, Estonian, Russian, and English are the ones our Baltic clients run daily), transcripts and summaries, caller history, the analytics dashboard, and model upgrades as they ship.
- Setup: free on the self-serve path (the first agent is generated from your website in minutes and calls you) and free for standard onboarding. Custom integration projects, such as connecting a booking system or a niche CRM, typically run €500 to €1,000 one time.
- What a licence means: one licence is one AI agent with its own phone number; included minutes pool across a team's licences.
- Your number: keep your existing number and forward it to the agent, or use the included one.
- Runaway protection: outbound dialing hard-stops at twice the included minutes, so a runaway campaign cannot produce an unbounded bill; inbound calls are never blocked.
All prices exclude VAT. There is no annual lock-in and no per-feature meter.
Common questions
Why do per-minute voice AI prices look so much cheaper than they are?
Because the headline number is rarely the whole bill. Developer platforms advertise the platform fee per minute; the language model, speech recognition, voice synthesis, and telephony are billed on top, and you still pay an engineer to build and maintain the agent. Bland's own comparison estimates a typical production stack at $0.13 to $0.30 per minute on Vapi and $0.11 to $0.25 on Retell once those are added. Managed vendors that bill per minute add a monthly subscription and often a four-figure setup fee around the per-minute rate.
At what call volume does a flat subscription beat per-minute pricing?
Against a typical European offer of a €149 subscription plus €0.39 per minute, a €249 flat plan with 1,000 minutes included is cheaper above roughly 250 minutes per month, which is about 85 average business calls. A company that currently misses a quarter of its inbound calls is almost always above that line.
Does Canarics charge a setup fee?
No. Self-serve setup is free: sign up, verify your number, and the first agent is generated from your website. Standard onboarding is also free. Custom integration projects, such as connecting a booking system or a niche CRM, typically run €500 to €1,000 one time.
What is included in the Canarics monthly price?
Everything the agent needs to run: the phone number and telephony, the AI conversation itself in more than 40 languages, transcripts, summaries, caller history, the analytics dashboard, and ongoing model upgrades. 1,000 minutes are included per licence, and most clients stay under that, so in practice the subscription is the whole bill; additional minutes are billed in 100-minute blocks at the same per-minute rate.
Why can Canarics be cheaper, and is it actually better?
The lower price is structural, not a discount. Onboarding is automated, the first agent is generated from your website, so there are no consultant hours to recoup through a setup fee. The stack is our own, built directly on carrier and cloud integrations rather than resold on top of a per-minute platform that takes its own margin. Are we actually cheaper? Against metered offers above roughly 250 minutes a month, and against any self-built stack once the engineering is counted, yes; below 250 minutes a metered offer wins, and a developer team that wants full control is not our buyer. Better is measurable where it matters to a business phone line: real conversations rather than announcements, mid-call language switching, memory of up to the last 20 conversations with each caller, 48.5% of conversations resolved end to end at a measured client, and EU data residency. Where we are not better: we are not a developer platform, and contact-center scale belongs on an enterprise contract.
+371 6433 9366, live AI demo line, no signup, standard rates. Prefer a walkthrough? Request a demo.