How Much Does a Custom AI Chatbot or Voice Agent Cost in 2026?

There are three ways to pay for a chatbot or voice agent: a monthly plan with a usage cap, a meter by the minute or by the resolved conversation, or a custom build billed as developer hours. Tidio Starter was $24.17 a month on November 4, 2026. A headline per-minute price leaves out lines, so sum all of them.

Got a voice agent quote in hand? Decode it below to see which of the five cost lines it includes. For the whole monthly bill, model tokens, voice minutes and workflow runs together, use the AI stack cost estimator.

Key facts

• Tidio Starter costs $24.17 a month, billing cycle unlabeled on the page (Tidio pricing, read Nov. 4, 2026).
• Intercom's Fin AI Agent starts at $0.99 per outcome (Intercom pricing, read Nov. 4, 2026).
• Vapi hosting is $0.05 a minute; models and telephony are billed on top (Vapi pricing, read Nov. 4, 2026).
• Kansas City software developers earn a median $60.09 an hour (BLS OEWS, May 2025).
• Only 3 of 29 ranking cost pages date their prices inline (my audit, 2026-11-04).

What a chatbot or voice agent is, and what it is not

A chatbot answers typed questions on your website or in an app. A voice agent answers phone calls, and it needs four parts to do that: speech to text, a language model, text to speech and a phone line. An AI receptionist is a packaged voice agent with one job, answering and booking calls, sold at a flat monthly price; I priced those against human answering services in my AI receptionist cost guide.

Three things it is not. It is not a person, so it will miss some calls and some questions. It is not a guarantee of resolved calls; per-minute platforms bill the minute whether or not the caller got an answer. And it is not free once you pass the cap, because every free tier I found has a ceiling. Not sure which kind you need? Read AI agent vs chatbot for small business.

The three ways to pay: plan, meter or custom build

A plan. A fixed monthly price with a cap on conversations, minutes or customers. Tidio Starter is $24.17 a month for 100 billable conversations. Chatbase Hobby is $40 a month, or $32 a month billed yearly, for 700 message credits. Rosie Professional is $49 a month for 250 call minutes. A plan fits you when your volume sits under the cap and you want one predictable bill.

A meter. You pay per unit used. Per minute on voice platforms: Vapi charges $0.05 a minute for hosting, then passes the model and telephony costs through at cost. Per outcome on chat: Intercom's Fin AI Agent is $0.99 per outcome. Per automated resolution: Zendesk bills its AI agents that way, and the per-resolution rate was not in the page text my script read on November 4, 2026. Per token at the API: OpenAI's gpt-realtime lists audio input at $32.00 and audio output at $64.00 per million tokens. A meter fits you when volume swings month to month, or when you want to pay for closed conversations instead of seats.

A custom build. Developer hours times an hourly rate, plus the meter above for usage. The Kansas City median wage for software developers is $60.09 an hour (BLS, May 2025). A build fits you only when you need integrations, data control or compliance that no plan offers.

Hybrids are common. Bland's Build plan is $0.12 a minute plus a $299 a month platform fee. Vapi Core is $29 a month on top of the per-minute hosting. ElevenLabs Agents sells minute bundles, Creator at $22 a month for 275 minutes, then $0.08 for each extra minute. When a price has two parts, write both down before you compare.

2026 price table, read November 4, 2026

Every figure below comes from the vendor's own pricing page, opened on November 4, 2026. Unlabeled billing cycles and yearly discounts are flagged in the notes.

Two of the 17 vendors in this table publish no price at all, Voiceflow and Synthflow. Two bill by the result: Intercom at $0.99 per outcome, and Zendesk per automated resolution with no dollar figure in the text I read. Prices move, which is why every row carries a date and a link you can click to re-check.

What a per-minute price really contains

A voice call runs through five billable lines: the phone line (telephony), speech to text, the language model, text to speech, and the platform that wires them together (hosting). Vapi's "$0.05/min Vapi hosting" and Telnyx's "$0.05 per minute" engine are the hosting line only. Vapi's page says models are pass-through at cost and telephony is charged by the provider. Telnyx bills language model tokens and carrier minutes separately, and its own all-in estimate on the same page is $0.06 a minute.

Worked example: the cheapest published components

Seven labeled inputs. The first two are assumptions; replace them with your own numbers.

• Input A: 120 calls a month (assumption).
• Input B: 3 minutes a call (assumption), so 360 minutes a month.
• Input C: telephony, Twilio, receive local calls, $0.0085 a minute.
• Input D: speech to text, Deepgram Nova-3 monolingual streaming, $0.0077 a minute (Deepgram's own pay-as-you-go rate; Vapi's page lists its Deepgram pass-through at $0.0095 to $0.0099).
• Input E: language model, Vapi's OpenAI pass-through band, $0.0077 to $0.0452 a minute.
• Input F: text to speech, Vapi's ElevenLabs pass-through band, $0.0146 to $0.0238 a minute.
• Input G: hosting, Vapi, $0.05 a minute.

Per minute, low: $0.0085 + $0.0077 + $0.0077 + $0.0146 + $0.05 = $0.0885. High: $0.0085 + $0.0077 + $0.0452 + $0.0238 + $0.05 = $0.1352.

Per 3-minute call: $0.2655 to $0.4056. At 360 minutes a month: $31.86 to $48.67, plus the $1.15 Twilio local number, is $33.01 to $49.82 a month before build labor, recording or extra concurrency.

The same 360 minutes on bundled platforms

Retell AI, with GPT 4.1 mini and platform voices: $0.055 + $0.0128 + $0.015 + $0.015 for telephony on a Retell number = $0.0978 a minute, or $35.21 a month, plus $2.00 a month for the Retell-provided number (read November 4, 2026).

Bland AI Start: $0.14 x 360 = $50.40, plus telephony at pass-through cost, which Bland bills separately.

ElevenLabs Creator: $22 covers 275 minutes; the other 85 minutes cost $0.08 each, $6.80; total $28.80, plus the language model at cost and the carrier, which bills you directly.

Three platforms, three different lines left out. Put your own quote through the decoder below and tick what it includes.

Decode a voice agent quote

Nothing you type here leaves your browser. Enter the quoted rate and your minutes, then tick which of the five lines the quote says it includes.

What it does not do: it fills any missing line with the lowest published rate I read on November 4, 2026, it does not check any vendor's claims, and it does not add build labor. For the full monthly bill, including model tokens and workflow runs, use the AI stack cost estimator.

Per resolution, per outcome and per seat: how chat bills

Chat vendors bill three ways, and the words matter. Intercom's Fin AI Agent costs $0.99 per outcome, charged once per conversation, with a minimum monthly commitment (Intercom, read October 31, 2026; the FAQ was collapsed when I read the page on November 4). An outcome counts when the customer confirms the issue is resolved, does not ask for more help after Fin responds, or Fin completes a workflow, including a handoff. Zendesk bills its AI agents per automated resolution on top of the per-agent seat, $19 per agent a month on Support Team paid yearly. Tidio sells a monthly plan capped by billable conversations: Free, Starter $24.17, Growth starting at $49.17, Plus starting at $300.

Chat example, with two labeled inputs. Input H: 150 website chats a month (assumption). Input I: 60% handled without a person (assumption; measure it in a trial). That is 90 outcomes x $0.99 = $89.10 a month at Intercom's published rate, before the minimum commitment and before seats for the people who take the other 60 chats.

Read the definition of "resolved" before you sign, because it sets your bill. "Does not ask for more help after Fin responds" counts a customer who gave up as an outcome. The percentage is yours to measure, in your own trial, on your own questions. The only published floor I saw is Tidio's Premium plan, which mentions pay per resolution with a "guaranteed 50% resolution rate" and no price (read November 4, 2026).

Two numbers to ignore. The "80% of tickets" line traces to a Gartner press release dated March 5, 2025, which predicts agentic AI "will autonomously resolve 80% of common customer service issues" by 2029; a forecast about common issues is not your rate. Chatbase's use-case page shows 80% "resolved without a human" on a single customer card with no method, period or sample.

What a custom build costs: hours times rate, plus usage

Price a build as labeled inputs, not a borrowed range: hours times a published rate, plus overhead, plus the usage meter above.

• Hours: assumption, 40 for a single-job agent, 120 for a build with two or three integrations.
• Rate: $60.09 an hour, the median hourly wage for software developers in the Kansas City metro, BLS OEWS May 2025 (12,160 employed, annual median $124,990; series OEUM002814000000015125208, read November 4, 2026).
• Overhead: 20%, a labeled assumption borrowed from my AI agent cost calculator.

40 hours x $60.09 = $2,403.60 in wages, x 1.2 = $2,884.32 with overhead. 120 hours x $60.09 = $7,210.80, x 1.2 = $8,652.96. Then add usage: $33.01 to $49.82 a month for the 360-minute voice example, or $89.10 for the 90-outcome chat example.

Why this beats the ranges you will see elsewhere. Agency pages print custom build figures such as $20,000 to $100,000 with no source, no hours and no rate, and the pages printing them sell builds. Hours times a public wage is a number you can argue with; a range is not. If you hire someone to scope and build it, my AI consultant cost guide shows what that time costs in Kansas City.

Maintenance is real, and I will not give you a percentage for it. What changes: models get deprecated and the replacement answers differently; carriers and platforms change rates; prompts need updates when your hours, services or prices change. Budget hours for those, at the same rate.

Blur Agentic, my business, scopes custom builds like these. I say so because I have a stake in that answer.

Hidden costs: setup, telephony, overage, concurrency, compliance

Setup. None of the 18 third-party vendor pricing pages I opened on November 4, 2026 shows a setup fee amount. Setup still costs you. It lives in your own hours, or in a platform fee like Bland Build's $299 a month or Vapi Pro's $999 a month minimum.

Telephony. Separate on Retell ($0.015 a minute on Retell numbers), Vapi (charged by the provider: Twilio inbound $0.008 a minute, Telnyx $0.0055), Bland (pass-through cost), ElevenLabs (the carrier bills you directly) and Telnyx (carrier from $0.0032 a minute). On Twilio's own US page, receiving a local call is $0.0085 a minute and a local number is $1.15 a month.

Overage and caps. ElevenLabs charges $0.08 for each extra minute and $0.16 a minute in burst when you exceed your concurrency limit. My AI Front Desk charges $0.25 a minute past its 200 minutes. Chatbase charges $99 a month to remove its Powered By label and $25 a month for an extra agent.

Concurrency. Retell includes 20 concurrent calls and charges $8.00 a month per extra. Vapi's base tier includes 4 and charges $10 a line a month. If your calls bunch up at lunch or after a storm, count how many overlap.

Recording and compliance. Twilio bills recording at $0.0025 a minute and recording transcription at $0.0500 a minute. Vapi's HIPAA add-on is $2,000 a month.

Free tiers are trials with a ceiling. ElevenLabs Free gives 15 call minutes. Retell gives $10 in credits, Vapi $5, Deepgram $200. Tidio Free allows 50 billable conversations a month. Chatbase has no free plan, only a 7-day trial. Quote the ceiling and the first paid step, then budget for the paid step. If this sits on top of other AI tools, my guide to what it costs to run an AI business covers the whole monthly bill.

How to price your own chatbot or voice agent, step by step

1. Count your real volume

Pull last month's phone bill and count inbound calls and total minutes. For chat, count the emails and form fills you answered by hand. Note how many came after hours.

Time: 30 minutes, my estimate.

Done when: you have three numbers on paper for one month: calls, minutes and chats.

2. Ask each vendor which lines the price includes, in writing

Send the same email to every vendor: does your price include telephony, speech to text, the language model, text to speech and hosting? Ask for the overage rate, the concurrency limit and the definition of a billable outcome. Keep the replies. For a builder, my 10 questions to ask an AI consultant covers the same ground.

Time: 1 hour to write and send, then the wait, my estimate.

Done when: every vendor on your list has answered all five lines in writing, or you have crossed it off.

3. Sum the lines with the decoder and the stack estimator

Run each quote through the decoder above, then enter the totals in the AI stack cost estimator to get one monthly bill with model tokens, voice minutes and workflow runs.

Time: 20 minutes, my estimate.

Done when: every option has one monthly number at your step 1 volume, with the same lines included.

4. Run a 7 to 14 day trial and measure your own resolution rate

Tidio, Chatbase and Rosie each list a 7-day trial. Route real calls or chats through it, then count how many ended without a person stepping in. That percentage, not a vendor's, sets your per-outcome bill. Do not pay for a year until you have it.

Time: 7 to 14 days of calls, plus 1 hour to count, my estimate.

Done when: you can write "X of Y calls handled without a person" with your own X and Y.

5. Publish your own prices and answers on a plain HTML page

A chatbot that answers "how much do you charge" should answer from a page that is public, so customers, ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews all see the same number. Put the price, the billing unit, the cap, any setup fee and an as-of date in plain text, add a short FAQ, and keep AI search crawlers allowed in robots.txt. None of this promises a citation. It improves the chance.

"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

Google Search Central, AI features documentation, read October 31, 2026

"Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers."

OpenAI, crawler documentation, read October 31, 2026

My check on November 4, 2026 found that 0 of 29 ranking cost pages say anything about getting a business found in AI search, and only 3 of 29 date their prices inline. The how-to is in how to get your Kansas City business in ChatGPT and Gemini, and the AI visibility scorecard checks your page.

Time: 2 hours to write and publish, my estimate.

Done when: your pricing page shows a price, a unit and a date in plain text, your robots.txt allows the crawlers you want, and your chatbot's price answer matches the page.

Our data: what 29 ranking pages and 18 vendor pricing pages show

I ran this on November 4, 2026: Google US page one (via Serper) for six queries ("how much does a chatbot cost", "ai voice agent cost", "how much does an ai phone agent cost", "ai chatbot pricing for small business", "custom chatbot development cost", "voice ai pricing per minute"), with Reddit, Quora, YouTube, forums and vendor pricing pages dropped, 31 unique pages loaded in headless Chrome by script, and 29 of them readable articles. The same day I opened 18 vendor pricing pages and checked whether each figure I recorded on October 31 still appeared in the page text. Keyword checks can miss rewording and over-match, so they count pages that use the terms, not quality; two checks, prices dated inline and advice on AI search, I did by hand on the saved text.

The files: ranking page checks (CSV) and vendor pricing page re-checks (CSV).

Three examples from the file. Elfsight's dated table says "as of February 2026", about nine months old on the day I ran this; the other two top pages, Crescendo at 2,143 words and Quickchat at 4,691 words, carry no dated prices, and none of the three names a Kansas City labor rate or sums per-minute components. Only 3 of 29 pages date their prices inline, and 0 of 29 mention Kansas City. On the vendor side, four of 18 pages (Tidio, Intercom, Zendesk, Deepgram) matched only in part after four days, with one recorded figure each gone from the page text, and Smith.ai's AI pricing page showed my script only a contact form.

The honest takeaway: most of what ranks for this query is undated, and vendor prices move inside a week, so a price without a date is already a guess. That includes this page after today, which is why every figure here carries November 4, 2026 and a link to re-check.

Mistakes that cost the most

1. Comparing headline per-minute prices

A "$0.05 a minute" quote is the hosting line. The cheapest published full stack sums to $0.0885 to $0.1352 a minute. $0.0885 less $0.05 is $0.0385, and 360 x $0.0385 = $13.86; $0.1352 less $0.05 is $0.0852, and 360 x $0.0852 = $30.67. That is $13.86 to $30.67 a month the quote left out.

Fix: ask which of the five lines are in the price, in writing, before you compare.

2. Paying per outcome without defining outcome

Example arithmetic: 90 outcomes x $0.99 = $89.10. If the vendor counts all 120 of your conversations as outcomes, including the ones where the customer just stopped replying, it is 120 x $0.99 = $118.80, a $29.70 difference on the same month.

Fix: get the definition of "resolved" in writing and check it against your own trial counts.

3. Buying a year before a trial

Chatbase Business is $400 a month billed yearly, $4,800 for the year, against $500 a month billed monthly, $6,000 over twelve months. The page states the $1,200 saving. A wrong fit locks in the $4,800. The same pattern sits behind every annual toggle in the table.

Fix: trial on the monthly price first, then switch to annual once your own numbers hold.

Plan vs per-minute platform vs custom build vs packaged receptionist

Disclosure first: I run Blur AI Solutions and Blur Agentic, which sell voice and chat automation, so I have a stake in this comparison. Setup and effort below are my estimates.

My pick for most small businesses: under about 1,000 minutes a month, usage cost is small next to setup and your own labor, so compare caps, overage and what is included, and start on a plan or a packaged service with a trial. Build custom only for integrations, data control or compliance that no plan offers.

How I weighted the options

Cost at your volume leads because it is the one number every headline rate hides, and the one you pay every month.

I run businesses that sell voice and chat automation, Blur AI Solutions and Blur Agentic, so I have a stake in this topic. I get nothing from any vendor named on this page. The decoder is free, asks for no email and sends nothing to me. This is not legal or financial advice.

Verdicts, by type of business

Who this is not for

Enterprises with thousands of calls a day. The rates here are pay-as-you-go floors. At that volume, negotiate: Retell, Vapi, Bland, Synthflow and ElevenLabs each list a custom or enterprise tier.

Businesses that want a human answering service. A person on the line is billed per call or per minute of human time, and the math is different. My AI receptionist cost guide prices human answering services next to AI.

Regulated calls or outbound calling. Medical, legal and lending calls carry recording, consent and disclosure rules, and outbound AI calls carry their own. Get a compliance review before you build or buy. This is not legal advice.

How I put this together

I opened and compared the top three ranking articles for "how much does a chatbot cost" and the two voice leaders on November 4, 2026.

• Crescendo (2,143 words) gives a quick summary block, pricing models and industry examples. It has no dated prices and no per-minute math.
• Elfsight (3,391 words) has a 16-row vendor table dated "as of February 2026", three scenarios and a FAQ with schema. It covers chat only, with no voice, telephony or build hours, and no links to vendor pricing pages.
• Quickchat (4,691 words) shows a credit formula with worked cases, a first-person test and an FAQ, updated October 30, 2026. Its build ranges are undated and unsourced, and voice gets no per-minute price.
• Aircall names the four voice lines and gives bands. It prices no chat and shows no component math.
• Famulor has a per-minute formula and three volume scenarios. Its inputs are marked approximate.

This page adds: 17 vendor rows, each with the pricing page linked and read on November 4, 2026; the per-minute component sum with seven labeled inputs; chat and voice priced on one page at the same volume; a build formula at the Kansas City rate; a decoder you can use without giving an email; and a section on getting your own pricing page cited by AI search, which 0 of 29 ranking pages cover. More about me on the about page.

Run your quote through the decoder, use the AI stack cost estimator for the monthly bill, or email [email protected] with your monthly calls, minutes or chats and I will price your exact setup. For an after-hours AI phone receptionist, Blur AI Solutions (my business) publishes its prices and has a live demo line on its site. Next read: AI receptionist cost in Kansas City.