AI Voice Agent Pricing: Vapi, Retell AI, Bland AI and Synthflow Cost Per Minute

How much does an AI voice agent cost per minute?

The honest band is $0.05 to $0.31 per connected minute, and where you land inside it depends far more on how the vendor slices the bill than on how good the agent is.

There are only two real models. The first is all-in: one rate that already contains the language model, speech to text and text to speech. Bland AI is the clean example at $0.11 to $0.14 a minute with, in its own words, no token charges. The second is unbundled: a thin platform fee plus the components at cost. Vapi is the clean example at $0.05 a minute for the platform and everything else passed through.

Retell AI is the useful one, because it publishes both views at once. Its headline is $0.07 to $0.31 a minute, and directly underneath it prints the stack that produces that range: $0.055 for Retell voice infrastructure, $0.16 for the language model on a GPT 5.5 example, $0.015 for text to speech and $0.015 for US Twilio telephony. Add those and a real call is about $0.245. The spread between $0.07 and $0.31 is not a discount you negotiate. It is which model you point the agent at.

Vapi vs Retell AI: which one is actually cheaper?

On the number each one advertises, Vapi wins by a mile: $0.05 against $0.07 to $0.31. On the number you pay, they finish close together, and Vapi can finish higher.

The reason is that the two figures are not the same kind of number. Vapi $0.05 is a platform fee. Speech to text, the language model and text to speech are billed at cost, and Vapi publishes no price for any of them. Retell $0.07 to $0.31 already contains those components. So the comparison only becomes real once you supply the missing Vapi components, and the only published US prices for them come from Retell own page.

Doing exactly that, and labelling it as our arithmetic rather than anyone published rate: take Retell published components for the language model ($0.16), text to speech ($0.015) and telephony ($0.015), add them to the Vapi $0.05 platform fee, and a comparable Vapi minute is around $0.24. A comparable Retell minute on the same stack is $0.245. The gap is half a cent.

Where they genuinely diverge is concurrency and control. Vapi includes 10 concurrent lines and charges $10 per line per month after that. Retell includes 20 and charges $8 per concurrency per month. At 50 lines that is $400 a month on Vapi against $240 on Retell. Vapi gets it back if you bring your own API keys, which drops the model cost to $0 on your Vapi invoice and moves it to your model provider invoice instead. That is a real saving for a team that already buys inference at volume, and no saving at all for a team that does not. If you are weighing both, our Vapi alternative and Retell AI alternative pages cover what each one asks you to build.

How much does Bland AI cost per month?

Bland AI publishes four tiers and is the easiest of the group to budget, because the per minute rate is all-in.

Start is $0.14 a minute with no platform fee, and includes 2 credits plus an inbound number that Bland values at $15 a month. Build is $0.12 a minute plus $299 a month. Scale is $0.11 a minute plus $499 a month. Enterprise is custom, contracted to volume, with on premise or VPC deployment, a BAA, SSO and data residency available.

The line that matters more than the price is the daily call cap. Start allows 100 calls a day, Build 2,000 and Scale 5,000. Concurrency is 10, 50 and 100. For an outbound team that is the binding constraint, not the rate: a 100 call daily ceiling is roughly what one human rep does before lunch, so Start is a pilot plan rather than a production one no matter how attractive the $0 platform fee looks. Transfers are billed separately at $0.05, $0.04 and $0.03 per transfer minute by tier, which matters if your agent hands warm calls to a human closer. We cover the build tradeoff on our Bland AI alternative page.

What does ElevenLabs Agents pricing actually buy you?

ElevenLabs Agents runs Free at $0, Starter $6, Creator $22, Pro $99, Scale $299 and Business $990 a month, with Enterprise custom. Included agent minutes are 15, 75, 275, 1,238, 3,738 and 12,375, and additional minutes are $0.080.

Divide each allowance into its plan fee and something interesting falls out. Starter is $6 for 75 minutes, which is $0.080 a minute. Creator is $22 for 275 minutes: $0.080. Pro is $99 for 1,238 minutes: $0.080. Scale is $299 for 3,738: $0.080. Business is $990 for 12,375: $0.080. Every paid tier prices its bundled minutes at exactly the overage rate.

So upgrading does not buy a cheaper minute. It buys concurrency, which goes 4, 6, 10, 20, 30 and 40 by tier, and it buys headroom before you hit burst pricing, which is $0.160 a minute, precisely double, for calls that exceed your concurrency cap. Two other things are worth pricing in before you compare $0.080 against anyone else: the language model and any telephony are billed separately on top, based on usage, so $0.080 is a voice cost rather than a call cost.

How much does Synthflow cost?

Synthflow no longer publishes a self serve price. Its pricing page states that enterprise contracts start at $30,000 annually, which is $2,500 a month committed before a single minute is dialed.

The page is explicit that the final figure is scoped around call volume, concurrency, telephony setup, integrations, security needs and launch support, and that the package includes custom concurrency planning, routing, enterprise SLA terms and implementation support. That is a genuine enterprise motion rather than a hidden price, and it puts Synthflow in a different bracket from the usage based platforms above it: at 5,000 minutes a month the Synthflow floor alone is more than twice what a comparable Bland AI bill comes to.

If you are evaluating it against lighter options, our Synthflow alternative page covers what the contract is buying.

The same 5,000 minutes, priced on every platform

Vendor pages are not comparable, so here is one scenario applied to all of them: 5,000 connected minutes a month, 20 concurrent lines, US telephony. Where a vendor passes components through at cost we use Retell published component prices as the reference stack, because they are the only published US figures available, and we label that arithmetic as ours.

Bland AI Build: $0.12 times 5,000 is $600, plus the $299 platform fee, is $899 a month. Concurrency of 50 is already included, and the model and voice are in the rate. Bland Start would be $700 with no platform fee, but its 10 concurrent lines and 100 call daily cap do not meet the scenario.

Retell AI on its own published example stack: $0.245 times 5,000 is $1,225 a month, with 20 concurrent calls free. Worth noting how much the model choice moves this: at the $0.07 bottom of the published range the same 5,000 minutes is $350, and at the $0.31 top it is $1,550.

Vapi Build: the platform fee is $0.05 times 5,000, or $250, plus 10 extra concurrent lines at $10 each, or $100. That is $350 before any model, voice or telephony cost at all. Adding the reference components at $0.19 a minute brings it to roughly $1,300 a month, or back down toward $350 if you already buy inference at volume and bring your own keys.

ElevenLabs Scale: $299 covers 3,738 minutes, and the remaining 1,262 at $0.080 adds $101, so about $400 a month for voice. Concurrency of 30 covers the scenario. Add the reference language model and telephony and the all-in figure is nearer $1,275.

Synthflow: $2,500 a month minimum, from the $30,000 annual floor, whatever the usage.

ColdCalls.ai Growth: $1,490 a month for 3 seats and about 6,000 connected minutes, which covers the scenario with headroom. There is no separate model bill, no telephony bill, no concurrency line and no build.

Per minute or per seat: what AI SDR pricing looks like next to a voice API

The platforms above sell you the ability to make a call. A different set of vendors sells you the outbound job, and they price per seat or per month instead of per minute, which makes a straight comparison misleading in both directions.

AiSDR is Solo $250, Explore $900 and Scale $2,500 a month, capped at 200, 800 and 2,500 AI researched contacts. Solo is a single user on monthly billing; Explore and Scale carry unlimited users on quarterly contracts. The important detail for anyone shopping this page is that AiSDR call steps run through an Aircall integration where a human dials, so the seat fee does not contain outbound calling in the sense the voice platforms mean it. Regie.ai has moved to credits: Free with 250 one time credits, Pro at $49 a month with 5,000 credits, Enterprise custom, and it does not publish a credit to minute conversion. 11x publishes nothing at all, and its pricing page has returned a 404 on every check since July 2026.

The comparison that actually decides a budget is total cost of ownership, not rate. A voice API bill of $900 to $1,300 a month is the platform only. On top of it sits prompt and flow engineering, a telephony account, number reputation management, DNC scrubbing, consent records, retention and the CRM write back, and somebody has to own all of it. We worked that arithmetic through in AI vs human SDR cost and priced the calling side in AI cold calling cost.

The line items that are not in any per minute rate

Four costs sit outside every rate on this page, and they are the ones that surprise a first invoice.

Concurrency. It is priced as a product, not a feature. Vapi is $10 per line per month past 10, Retell is $8 per concurrency per month past 20, Bland bundles it into the tier, and ElevenLabs charges $0.160 a minute, double its standard rate, for calls that burst past your cap. At 50 lines that alone is $400 a month on Vapi and $240 on Retell.

Numbers and reputation. Retell charges $10 per verified phone number per month and $0.10 per branded outbound call. On a US outbound program that rotates numbers to protect answer rates, that is a recurring line, not a one off. The same problem in the dialer world is priced separately too, which we broke down in auto dialer pricing.

Compliance and data handling. Vapi prices HIPAA at $2,000 a month and Zero Data Retention at $1,000 a month, and keeps Build call history for only 14 days. If you are in a regulated vertical, that is a real number to add before comparing per minute rates. It is also worth knowing what US telemarketing law expects of the recordings and dispositions you keep: 16 CFR 310.5(a) requires per call records including the script used, the caller ID transmitted and the disposition of each call, kept for five years, which a 14 day history does not satisfy on its own. We cover that on TCPA compliant AI calling.

Volume caps. Bland caps calls at 100, 2,000 and 5,000 a day by tier. ElevenLabs caps concurrency by tier. AiSDR caps researched contacts at 200, 800 and 2,500 a month. A cap has no price, so it never appears in a comparison table, and it is frequently the thing that forces the upgrade.

Which pricing model fits which kind of outbound team?

If you have engineers and you already buy inference at volume, Vapi is the cheapest floor available, because bringing your own API keys moves the largest component off the Vapi invoice entirely. You are paying $0.05 a minute plus $10 a line for orchestration and buying everything else yourself.

If you want one predictable number and no model shopping, Bland AI is the simplest bill on this page. All-in per minute, no token charges, concurrency bundled. Watch the daily call cap rather than the rate.

If you want to see exactly what you are paying for, Retell AI is the most transparent, and it is the only vendor here that publishes its component stack. That transparency is also what makes the $0.07 to $0.31 range look alarming when it is really just model choice.

If outbound calling is the job rather than the project, per minute pricing is measuring the wrong thing. Nobody buys minutes; they buy booked meetings. That is why our own pricing is a flat managed fee with connected minutes bundled and no per meeting cut, and why we compare tools by who actually talks on the call in best AI cold calling software. If you are earlier in the search, AI voice agent for sales covers what these platforms do before you price them, and outbound call center pricing covers the seat based contact center side of the market.