Building
AI receptionist cost in 2026: per call, per minute, and what each part of a call costs
What an AI receptionist costs in 2026: current prices from 7 services and 5 voice platforms, the cost of each part of a call, and examples at 3 volumes.
In short
For a small business taking about 300 calls a month, a packaged AI receptionist service costs $129–$500 a month at list prices on November 4, 2026, or $0.43–$1.67 a call. Voice platforms you build on charge $0.07–$0.31 a minute, and the parts underneath (phone line, speech recognition, voice and AI model) cost about $0.03 a minute. A human answering service costs about $7 a call. What you pay depends most on what the provider meters: calls, minutes or unique callers.
Key takeaways
- At 300 calls a month, packaged AI receptionist services list at $129–$500 a month; a human answering service at about $2,100.
- The parts of a call cost about $0.03 a minute, and the AI model is the smallest line: about a tenth of a cent with a small, cached model.
- The meter matters more than the price: per-caller and per-call plans favour long calls, per-minute plans favour short ones.
- Check what’s billed: spam, short calls, hold time, transfers and going over your plan are handled differently by every provider.
- Voice platforms cost $0.07–$0.31 a minute in usage, plus the build, testing and upkeep that packaged services include.
How much does an AI receptionist cost?
For a small business taking about 300 calls a month, a packaged AI receptionist service costs $129–$500 a month at list prices, or $0.43–$1.67 a call. Building on a voice platform costs about $70–$110 a month in usage at the same volume, before the build itself. The parts underneath (the phone line, speech recognition, the voice and the AI model) cost about $0.03 a minute. A human answering service costs about $7 a call.
We checked every price in this article on each provider’s own US pricing page on November 4, 2026. Prices change often, so the method at the end works with whatever you find on the day.
| Kind | Examples | Per month | Per call |
|---|---|---|---|
| Packaged AI receptionist service | Goodcall, Rosie, Upfirst, My AI Front Desk, Dialzara, Smith.ai | $129–$500 | $0.43–$1.67 |
| Voice platform you build on | Vapi, Retell, ElevenLabs Agents, Bland | $68–$111 in usage | $0.23–$0.37 |
| The parts, run yourself | Twilio, Deepgram, GPT-5 mini | About $26 in usage | $0.09 |
| Human answering service | Smith.ai receptionists | $2,100 | $7.00 |
Platform and parts figures are usage only. They leave out building the agent, testing it on real calls and keeping it up to date, which packaged services include in their price.
How are AI receptionists priced?
Every provider meters something different, and the meter matters more than the headline price:
- Per call: Smith.ai’s AI receptionist and Upfirst. Call length doesn’t change the price, so long calls are good value.
- Per minute: Rosie, My AI Front Desk, Dialzara and every voice platform. Short calls are cheap; long intake calls add up.
- Per unique caller: Goodcall bills each phone number once a month, with unlimited calls and minutes. Repeat callers cost nothing extra.
- Plans with included minutes: ElevenLabs Agents, and Rosie, which has no overage rate at all.
The same business can be cheapest on one meter and most expensive on another. In our law-firm example below, 400 six-minute intake calls cost $129 a month on Goodcall’s per-caller plan, $717 on Smith.ai’s per-call plans and $839 on Dialzara’s per-minute plans.
AI receptionist service prices in 2026
| Service | Plans | Included | Beyond the plan |
|---|---|---|---|
| Goodcall | $79, $129, $299 a month | 100, 500 or 2,000 unique callers; unlimited minutes | 79¢, 26¢ or 15¢ per extra caller |
| Upfirst | $24.95–$299 a month | 30–600 calls | $1.50 down to 70¢ a call |
| Rosie | $49, $149, $299 a month | 250, 1,000 or 2,000 minutes | No overage rate: you move up a plan |
| My AI Front Desk | $99 a month | 200 minutes | $0.25 a minute |
| Dialzara | $29–$349 a month | 60–1,000 minutes | $0.48 down to $0.35 a minute |
| Smith.ai AI receptionist | $150–$800 a month | 75–500 calls | $2.50 down to $2.10 a call |
| Synthflow | Not published | Annual enterprise agreements only | Not published |
Annual billing lowers some of these: Rosie gives two months free, My AI Front Desk drops to $79 a month, and Upfirst takes 20% off.
What you get on each plan varies as much as the price:
- Booking into your calendar: Rosie from its $149 plan (the $49 plan texts the caller a booking link), Dialzara from $99, and Smith.ai on every plan.
- Transferring to a person: Rosie from $149. Dialzara’s $29 plan only does blind transfers; warm transfers start at $99.
- Spanish: on every Rosie plan, and on Dialzara from $99.
- Your CRM: Goodcall connects natively to Google Calendar, Google Sheets and HighLevel, and to other tools through Zapier. Rosie connects through Zapier only. Smith.ai lists native integrations such as Clio.
What do voice platforms cost?
If you build your own receptionist, or have one built, it usually runs on a voice platform that handles the call and bills by the minute.
| Platform | Usage price | Billed on top | Concurrent calls included |
|---|---|---|---|
| Vapi | $0.05 a minute | Speech, voice, model and phone line at provider cost; its estimator puts 1,000 minutes at $82–$129 before the phone line | 4, then $10 a line a month |
| Retell | $0.07–$0.31 a minute: $0.055 infrastructure, $0.015–$0.10 voice, plus the model | US phone line $0.015 a minute, number $2 a month; knowledge base $0.005 a minute | 20, then $8 a month each |
| ElevenLabs Agents | $99, $299 or $990 a month for 1,238–12,375 minutes, then $0.08 a minute | The model; the phone line from your carrier | 20, 30 or 40 |
| Bland | $0.14 a minute, or $299 a month plus $0.12 | The phone line | 10, or 50 |
| Deepgram Voice Agent API | $0.075 a minute for speech, model and voice together | The phone line | Not listed |
Retell bills by the second, including silence and hold; after a transfer, only the phone line keeps running.
What does one minute of a call cost, part by part?
Under every product above, an AI receptionist is four services on one call:
- A phone line that answers and streams the caller’s audio to your agent.
- Speech to text that transcribes the caller as they talk.
- A language model that reads your instructions and the relevant passages from your documents, and decides what to say.
- Text to speech that says it.
The phone line and speech to text run for the whole call. The voice bills by the character, so it only costs while the agent speaks. The model bills by the token, every time the agent replies. Here is that arithmetic as code, with the assumptions written down:
// What one minute of an inbound call costs, part by part, at list prices.
type Prices = {
phonePerMin: number; // carrier inbound + streaming the audio to your agent
sttPerMin: number; // streaming speech to text
ttsPer1kChars: number; // the agent's voice
llmInPerM: number; // per million tokens
llmCachedPerM: number;
llmOutPerM: number;
};
type Call = {
agentTalkShare: number; // share of the call the agent is speaking
wordsPerMin: number;
charsPerWord: number;
turnsPerMin: number; // agent replies per minute
cachedTokens: number; // instructions and passages re-sent each turn
freshTokens: number; // new conversation and new passages each turn
outTokens: number; // tokens the model writes per reply
};
export function costPerMinute(p: Prices, c: Call) {
const phone = p.phonePerMin;
const stt = p.sttPerMin;
const tts = ((c.agentTalkShare * c.wordsPerMin * c.charsPerWord) / 1000) * p.ttsPer1kChars;
const llm =
(c.turnsPerMin *
(c.cachedTokens * p.llmCachedPerM + c.freshTokens * p.llmInPerM + c.outTokens * p.llmOutPerM)) /
1e6;
return { phone, stt, tts, llm, total: phone + stt + tts + llm };
}
const prices: Prices = {
phonePerMin: 0.0085 + 0.0044, // Twilio local inbound + Media Streams
sttPerMin: 0.0077, // Deepgram Nova-3, pay as you go
ttsPer1kChars: 0.03, // Deepgram Aura-2
llmInPerM: 0.25, // GPT-5 mini
llmCachedPerM: 0.025,
llmOutPerM: 2.0,
};
const call: Call = {
agentTalkShare: 0.45, wordsPerMin: 150, charsPerWord: 5.5,
turnsPerMin: 3, cachedTokens: 3000, freshTokens: 800, outTokens: 40,
};
const m = costPerMinute(prices, call);
console.log(Object.fromEntries(Object.entries(m).map(([k, v]) => [k, v.toFixed(4)])));
// 300 calls a month at 2.5 minutes, plus one local number at $1.15
console.log((300 * 2.5 * m.total + 1.15).toFixed(2));We type-checked and ran it. The assumptions are an illustration, not a measurement: the agent speaks 45% of the time at 150 words a minute and replies three times a minute. Each reply re-sends 3,000 tokens of instructions and passages, which the provider caches, plus 800 new tokens, and writes 40.
| Part | Priced with | List price | Per minute of call |
|---|---|---|---|
| Phone line | Twilio local number, with Media Streams | $0.0085 + $0.0044 a minute | $0.0129 |
| Speech to text | Deepgram Nova-3, streaming | $0.0077 a minute | $0.0077 |
| Voice | Deepgram Aura-2 | $0.030 per 1,000 characters | $0.0111 |
| AI model | OpenAI GPT-5 mini | $0.25 in, $0.025 cached, $2.00 out per million tokens | $0.0011 |
| Total | $0.0328 |
At 300 calls of 2.5 minutes, that is $25.75 a month including the number. What the numbers show:
- The model is the smallest line. Because the instructions and passages are cached between turns, each reply re-reads them at a tenth of the input price. Claude Haiku 5.5 brings the model line to $0.0004; a larger model such as GPT-5.4 mini, to $0.0030. Reasoning models bill their thinking as output tokens, so a voice agent runs them at a low reasoning setting, which also keeps replies fast.
- The phone line and speech are most of the bill. The voice you choose moves the total more than the model does: ElevenLabs Flash at $0.04 per 1,000 characters makes the voice line $0.0149.
- Toll-free numbers cost more to answer. Twilio charges $0.022 a minute to receive a call on a toll-free number, against $0.0085 on a local one.
- Speech-to-speech models cost more. Retell lists OpenAI’s GPT Realtime at $0.345 a minute and GPT Realtime mini at $0.07, against about $0.02 for speech to text, model and voice in the table above.
So why do platforms charge $0.07–$0.31 a minute when the parts cost $0.03? The difference pays for the hard part of a voice agent, which appears on no line above: telling when the caller has finished speaking, stopping mid-sentence when they interrupt, answering in under a second, and transferring cleanly. Running the parts yourself means building that, plus the server that runs your agent.
What does an AI receptionist cost at your call volume?
Three illustrations, each priced at the cheapest list-price plan for its volume, with overage where it applies:
- Home services: 300 calls a month averaging 2.5 minutes (750 minutes), from 250 unique callers.
- Law firm intake: 400 calls averaging 6 minutes (2,400 minutes), from 350 unique callers.
- Property management: 1,500 calls averaging 3 minutes (4,500 minutes), from 1,000 unique callers.
| Option | Home services | Law firm intake | Property management |
|---|---|---|---|
| Goodcall | $129 | $129 | $259 |
| Rosie | $149 | Custom plan | Custom plan |
| Upfirst | $160 | $235 | $929 |
| My AI Front Desk | $237 | $649 | $1,174 |
| Dialzara | $299 | $839 | $1,574 |
| Smith.ai AI receptionist | $500 | $717 | $2,900 |
| Vapi, with a Twilio line | $68–$103 | $216–$329 | $405–$617 |
| Retell, with GPT-5 mini and a knowledge base | $76 | $237 | $443 |
| ElevenLabs Agents, with a Twilio line | $109 | $221 | $413 |
| Bland, with a Twilio line | $111 | $356 | $668 |
| The parts, run yourself | $26 | $80 | $149 |
| Smith.ai human receptionists | $2,100 | $2,950 | $12,300 |
“Custom plan” means the volume is over the largest published plan. Smith.ai and Upfirst quote more than 1,000 calls a month as enterprise plans, so their property-management figures are the largest published plan plus overage. ElevenLabs Agents includes $0.003 a minute for the model, the figure in its own calculator.
Three patterns hold across the table:
- Per-caller pricing wins on long calls and repeat callers. Goodcall stays at $129 for the law firm because it never meters minutes.
- Per-minute pricing punishes long calls. Dialzara is $1.00 a call at 2.5 minutes and $2.10 at 6.
- Platforms get cheaper than most services as volume grows, but only on usage. At 1,500 calls a month, Upfirst, My AI Front Desk, Dialzara and Smith.ai cost about $500–$2,500 a month more than Retell’s usage. That difference is what pays for a build. Goodcall is the exception: below every platform’s usage at the two larger volumes.
The code for the plan side, so you can run your own volume:
type Plan = {
name: string;
monthly: number;
included: number;
unit: 'call' | 'minute' | 'caller'; // what the plan meters
overage: number | null; // null: no published rate past the allowance
};
type Usage = { calls: number; minutes: number; callers: number };
export function planCost(plan: Plan, u: Usage): number {
const used = { call: u.calls, minute: u.calls * u.minutes, caller: u.callers }[plan.unit];
const extra = Math.max(0, used - plan.included);
if (extra === 0) return plan.monthly;
return plan.overage === null ? Infinity : plan.monthly + extra * plan.overage;
}
// The cheapest published plan for your volume.
export function cheapest(plans: Plan[], u: Usage) {
return plans
.map((plan) => ({ plan: plan.name, cost: planCost(plan, u) }))
.sort((a, b) => a.cost - b.cost)[0];
}
const lawFirm: Usage = { calls: 400, minutes: 6, callers: 350 };
console.log(cheapest([
{ name: 'Dialzara Business Plus', monthly: 199, included: 500, unit: 'minute', overage: 0.40 },
{ name: 'Dialzara Business Elite', monthly: 349, included: 1000, unit: 'minute', overage: 0.35 },
], lawFirm));
// { plan: 'Dialzara Business Elite', cost: 839 } How does that compare with a human receptionist?
The median US receptionist earned $18.27 an hour, or $38,010 a year, in the Bureau of Labor Statistics’ May 2025 figures. Benefits make up 31.3% of total compensation for office and administrative support workers in private industry. Together that puts a median receptionist at roughly $26.60 an hour to the employer, or about $55,000 a year. That is our estimate, combining a median wage with an average benefits share.
That is about $4,600 a month for 40 of the week’s 168 hours. Evenings, weekends, lunch breaks and sick days still need cover. Every AI option in the table above costs less than that at all three volumes. Smith.ai’s human service costs less at the two smaller volumes, and more at 1,500 calls.
A person still handles what an agent can’t: an upset caller, an unusual request, a judgement call. Plan for someone to transfer to, and check that the plan you pick can transfer.
How do you estimate your own cost?
- Export the last 90 days of call logs from your phone system or carrier.
- Work out calls a month, the average call length, the number of unique callers, and the share that is spam or under 15 seconds.
- Count how many calls arrive at once in your busiest hour.
- Price each option on its own meter at the cheapest plan that fits, with overage.
- Add what applies from the section above: numbers, recording, and the plan you need for booking, transfers or Spanish.
- Compare the total with what you pay now for the same hours of cover, including the calls that go to voicemail.
Should you buy a service or build your own?
Buy a packaged service if your calls are standard (hours, prices, messages, simple bookings) and one of the plans above has the features you need. Build on a platform if the agent must answer in depth from your own documents, work with your own systems, or handle volume where per-call fees outrun usage. Our AI receptionist guide has the full comparison, and the US rules on AI calls and recording.
We build the second kind: an agent that answers from your FAQs and documents goes live in three days, and one that books or looks things up takes about a week. See AI voice agents and receptionists. If a packaged service fits your calls, we’ll tell you.
Sources
- Goodcall — Pricing
- Goodcall — FAQ
- Upfirst — Pricing
- Rosie — Pricing
- My AI Front Desk — Pricing
- Dialzara — Pricing
- Smith.ai — AI receptionist pricing
- Smith.ai — Virtual receptionist pricing
- Synthflow — Pricing
- Vapi — Pricing
- Retell — Pricing
- ElevenLabs — Agents pricing
- ElevenLabs — API pricing
- Bland — Pricing
- Deepgram — Pricing
- Twilio — US voice pricing
- OpenAI — API pricing
- Anthropic — Claude pricing
- BLS — OEWS, Receptionists and information clerks (May 2025)
- BLS — Employer costs for employee compensation, table 4 (June 2026)