Voice AI Pricing

What a Billed Minute Contains, the Seven Rules That Change the Bill, and a 1,200-Minute Month Priced on Eight Vendors

Quick answer: Voice AI is priced by the minute almost everywhere, and the minute is a bundle. Retell's pricing page, read 15 September 2026, itemizes an example minute at $0.11: "LLM Cost $0.04/min, Retell Voice Infra $0.055/min, TTS Cost $0.015/min, Telephony Cost $0.00/min," the last because the example assumes you bring your own phone line. (That is the page's printed text; the calculator, once it runs in a browser, preselects GPT 4.1 at $0.045 and shows $0.115. We use the printed $0.11 as the reference minute throughout, with a $0.04 model such as GPT 5.1, and the half-cent is inside every rounding on this page.) Vapi's calculator for 1,000 minutes shows a $50 hosting fee, $10 of transcription, $8 to $45 of language model, $15 to $24 of voice, and an "Estimated Total $82–$129/mo." Those two pages agree on the structure even where they differ on the numbers: a phone line, a transcriber, a model, a voice, and a platform fee, each with its own rate. The rate is the smaller half of the bill. The larger half is the rules: Twilio rounds partial minutes up, Retell bills to the second but bills silence, ElevenLabs discounts silence by 95 percent and measures the call from connection rather than from hello, Retell stops its fee at transfer but keeps billing the line, and swapping the model can move the model's share of the minute from $0.0016 to $0.32 on the same platform. The worked month at the end puts 300 four-minute calls on eight vendors, and the spread is from about $11 (the phone line alone) to about $350 (Chatbase's credits at Standard), with most managed platforms landing between $100 and $200 before you touch the model.

This guide owns the decomposition of a billed voice minute, the billing rules that change it, and the worked month across our catalog. Whether voice is worth adding at all is the voice and chat hybrid guide; how to evaluate a vendor before signing is the voice bot buyer's guide; the per-call and per-caller alternatives that packaged receptionist products argue for are the AI receptionist entry; the second meter, lines held open at once, is the concurrent calls entry.

The five layers inside a minute

A voice bot is a pipeline, and each stage of the pipeline is a company with a price list. The table gives the published rate for each layer from a vendor that sells it on its own, so you can see what a platform is bundling when it quotes one number.

One caveat on the model row, and it is the best evidence on this page for why every figure here carries a date. We first read Retell's component table on 15 September 2026 and re-read it on 17 September while checking this guide. In two days three rows had been re-priced and three new rows had appeared. Re-priced: GPT 5 nano from $0.003 to $0.0016 a minute, GPT 4.1 nano from $0.004 to $0.0032, Claude 5 Sonnet from $0.08 to $0.064. New: GPT 6 Astra at $0.32 standard and $0.64 fast, which displaces GPT 5.5 as the most expensive standard-tier model on the page, plus GPT 5.6 Terra at $0.064 and GPT 5.6 Luna at $0.0064. Rows we quote elsewhere (Voice Infra $0.055, platform voices $0.015, ElevenLabs voices $0.040, GPT 5.5, GPT 5.4, GPT 4.1, GPT 5.1, Claude 4.5 haiku, the add-ons and the concurrency prices) were identical in both reads. The model row above and the model figures in this guide are the 17 September read; the $0.11 reference minute, whose $0.04 model rate is unchanged, is unaffected.

Two things follow from the table. The first is that the platform fee and the telephony are the stable parts of the minute, at roughly five cents and one cent respectively, and the model is the volatile part: on Retell's own list the cheapest model is $0.0016 a minute and the most expensive standard-tier model is $0.32, a factor of 200 on the same page, with everything else in the minute unchanged. The second is that a "speech-to-speech" model, which listens and talks without separate transcription and synthesis stages, is priced as a replacement for two layers: Retell lists GPT Realtime at $0.345 a minute and GPT Realtime mini at $0.07, and OpenAI's own pricing page lists its gpt-live-1 voice sessions at "$0.05" per minute, "billed per second, without rounding up to a whole minute," with "Backend model and tool usage … charged separately." Whether that is cheaper than the four-stage pipeline depends entirely on which model the pipeline would otherwise use; against a nano model it is not, against a frontier model it can be.

The seven rules that change the bill

The rate on the pricing page is a ceiling on a minute. What the minute is, and which minutes count, is decided by rules that sit in FAQs and documentation pages. These seven moved the arithmetic below more than any rate did.

1. Rounding. Twilio's help article on voice cost says "any partial minutes (under 60 seconds) are rounded up to the next full minute," and works the example: 61 seconds on a US toll-free number "would be charged $0.044 (the 61 seconds is rounded up to 2 minutes)." Retell's FAQ takes the opposite position: "Each call is tracked to the nearest second. At the end of each billing cycle, we charge based on your total accumulated minutes. No rounding up per call, no inflated bills." OpenAI bills gpt-live-1 "per second, without rounding up to a whole minute." ElevenLabs measures something different again: "The length of the call is measured based on the connection duration. This includes the time from when you begin the call, to when you end the call or the window is closed. This is why the call itself may be shorter than the duration you are charged for." For a business whose calls average four minutes and ten seconds, Twilio's rule bills five minutes of telephony per call, a 20 percent uplift on that layer; on a platform that tracks to the second, it bills 4.17. The layer that rounds is usually the cheapest layer, which is why the rule matters less than it sounds and more than zero.

2. Silence. A caller who says nothing is still on the line, and the transcriber is still listening. Retell's FAQ, asked whether you are charged during silence or hold, answers: "Yes. Billing covers the entire duration of the call because the speech-to-text engine remains active and listening throughout, even during silence." ElevenLabs' help article says voice-only calls carry "a 95% discount for periods of silence longer than 10 seconds." Chatbase's documentation gives you a setting instead of a discount: "End conversation after silence … Defaults to 300 seconds," and a "Max call duration … The default is 15 minutes." Our reading: a silence timeout is a billing setting wherever the platform has no silence discount, and Chatbase's 300-second default is five minutes of a stalled session charged at 6 credits a minute. It is the first setting to change on a new deployment, and Retell, which bills to the second and does not discount silence, caps a call at one hour by default rather than setting a silence timeout.

3. Transfers. A call handed to a human is not free, but it is cheaper. Retell: "Once the call is transferred, the AI voice agent fee stops. Only the telephony fee continues for the remainder of the transferred call." On Twilio the transferred call is two legs, and its help article says "prices are charged based on each leg of the call": the inbound leg to your number and the outbound leg to the phone you forwarded to, each at its own rate. A warm transfer, where the bot holds the caller while it reaches the human, is billed at the full agent rate until the human picks up, and the caller occupies one of your concurrent lines throughout.

4. Unanswered calls, failed calls, voicemail. Retell: "You are only charged for connected calls. If a call fails to connect, you won't be billed. For calls that reach voicemail, billing applies only for the duration the AI agent is active on the line." That last clause is the one to design around on outbound: an agent that talks to a voicemail greeting for forty seconds before deciding it is a voicemail is billed for forty seconds. Twilio sells answering machine detection at "$0.0075 / call," and Retell charges "+$0.005/dial" for batch calls, so the detection costs less than a minute of the agent's time and is worth turning on before any campaign.

5. The model. This is the lever that dwarfs the others, and it is the one the headline rate hides. On Retell the standard-tier rows run from $0.0016 a minute (GPT 5 nano) through $0.04 (GPT 5.1 and GPT 5, the model rate in the page's printed $0.11 example) and $0.045 (GPT 4.1, which the calculator preselects when it runs) to $0.32 (GPT 6 Astra), and the fast tier doubles the top row to $0.64. On Vapi and ElevenLabs the model is passed through at the model vendor's price, which Vapi's page states as "Models pass-through at cost" and ElevenLabs as "Based on usage and varies by model"; Deepgram's agent API takes $0.010 a minute off its bundled rate if you bring your own voice and $0.025 if you bring your own model and voice. For a 1,200-minute month on Retell, the difference between the nano model and GPT 6 Astra is $382, more than three times the entire rest of the bill. The voice bot buyer's guide makes the case that latency, not model size, is what a caller notices; the pricing pages make the same case in dollars.

6. Concurrency. Minutes measure the month; lines measure the peak. Retell includes 20 lines (its FAQ says per account, its documentation says the quota is per workspace) and charges $8 a month for each additional one; Vapi includes 4 on usage-only billing and 10 on its $29 Core package, with extra lines at $10 a month; ElevenLabs sets the number by plan (4 on Free, 20 on Pro) and bills calls over the limit at "$0.160" a minute against a standard "$0.080"; Chatbase gives 10 on Standard and 20 on Pro. Retell's burst option adds "$0.10/min" to "the entire duration of any call that started while in burst mode." The concurrent calls entry has the sizing arithmetic and what each vendor does to the caller when the lines are full; the point here is that a business with a Monday-morning rush buys its lines for that hour and pays for them all month.

7. Add-ons billed by the minute. The per-minute rate is the base of a stack of per-minute options. Retell: knowledge base "+$0.005/minute," advanced denoising "+0.005/min," PII removal "+0.01/min," and AI quality assurance "$0.10/min" after the first 100 minutes, which is almost a second agent's worth of cost for grading the first one. Twilio: call recording "$0.0025 / min" plus storage, Media Streams "$0.0044 / min," Conversation Relay "$0.07 / min." Deepgram: redaction "$0.0020/min," keyterm prompting "$0.0013/min." None is large; together, on a call with a knowledge base, denoising, PII removal and recording, they add about two cents to an eleven-cent minute, and QA adds ten.

How the five reviewed platforms with a voice surface bill a minute

Our voice AI ranking has five platforms with a first-party voice channel. Their pricing pages, read 15 September 2026, bill it five different ways.

The Chatbase row deserves a second look because it is the only one where the unit converts cleanly to a per-minute price. Six credits a minute at the Standard plan's implied rate of $0.0375 a credit ($150 for 4,000) is $0.225 a minute before any model credits; at the recharge rate of $0.04 a credit it is $0.24. That is roughly double Retell's default all-in minute, and Chatbase is not selling the same thing: the credits also buy a helpdesk, a chat widget, email and the integrations its review scores it on. But a buyer who wants voice alone should see the number, and the pricing page does not print it; the documentation does.

The same 300-call month on eight vendors

The scenario is the one our AI receptionist entry used for its per-call comparison, so the two pages can be read together: 300 inbound calls a month, four minutes each, 1,200 minutes, one US local number, no outbound. Monthly billing where a choice exists. Every line is our arithmetic on the vendor's published numbers; the vendor did not total these for us, except where noted.

Three things stand out from that column. The managed platforms that publish a rate cluster between $100 and $200 for this month once the line is included, and the difference between them is smaller than the difference a model choice makes inside any one of them. The cheapest total belongs to an API you would have to build a product on top of, which is the same finding our receptionist entry reached from the other direction: the cheapest number on the table is not a finished product. And the most expensive total is a credits system whose per-minute price is printed only in its documentation, which is not an argument against the platform but is an argument for reading the documentation before the pricing page.

What to ask before the pilot

The rates change; the questions do not. Ask each vendor, in writing, and put the answers next to the arithmetic above.

Which second of the call does billing start on: the ring, the connection, or the first word? ElevenLabs bills from connection; the others do not say. What is the smallest billed unit: a second (Retell, OpenAI), a minute (Twilio), or a credit that maps to a minute (Chatbase)? Is silence billed, discounted, or does it end the call, and after how long? When the bot transfers, what stops and what continues? Which model is the quoted rate assuming, and what is the per-minute difference to the one you would actually run? How many lines are included, what does the next one cost, and what does a caller hear when they are all busy? And is the rate on the page promotional, as Deepgram's streaming rate is labelled, or the regular price?

Then run the month three ways, as the voice bot buyer's guide recommends: at your real volume, at double, and at half. Voice is the channel where the demo minute and the production minute diverge fastest, because a production call is longer, includes silence, and sometimes ends in a transfer, and none of those is in the demo. The reduce chatbot costs guide has the chat-side version of this discipline; the chatbot ROI guide is where the minute cost meets what the call was worth.

What our fifteen reviews record

Searched 15 September 2026, case-insensitively, across the fifteen files matched by sample-reviews/*-review.md. per-minute or per minute appears in 2: the Voiceflow review, which records that voice deployments "require a separate Twilio account, which has its own per-minute and per-message rates," and the Wati review, where the phrase is about WhatsApp message throughput. telephony appears in 2 (Chatbase and Voiceflow); the Chatbase review places "Voice/Telephony/API access" on the $150 Standard tier and notes "We did not stress-test voice latency." concurren appears in 2, neither about calls. None of the fifteen protocols recorded a voice minute's cost from an invoice, timed a transfer against a bill, or placed simultaneous calls against a plan's limit. That is a gap in the protocol, recorded here; it is the reason every figure on this page is quoted from a vendor rather than measured by us. Corrections to editorial@chatbotscape.com.

FAQ

How much does an AI voice agent cost per minute?

On the platforms that publish a bundled rate, roughly $0.07 to $0.31 a minute before telephony, which is the span Retell prints for itself, with the language model as the main variable; a speech-to-speech model sits above it (Retell lists GPT Realtime at $0.345). Retell's printed example is $0.11 (a $0.04 model, $0.055 platform, $0.015 voice, your own line; its rendered calculator preselects GPT 4.1 and shows $0.115); Vapi's estimate for 1,000 minutes is $82 to $129; Deepgram's bundled agent API is $0.075; ElevenLabs is $0.08 for minutes beyond a plan's allowance plus the model at cost. Telephony adds about a cent a minute on a US local number.

What is included in a voice AI minute?

Five layers: the phone line, the transcriber that turns speech into text, the language model that decides the reply, the voice that speaks it, and the platform that runs the turn-taking, the flow and the logs. Some vendors bundle all five (Deepgram's agent API bundles four and leaves the line to you), some pass the model and the line through at cost (Vapi, ElevenLabs), and some sell credits that map to minutes (Chatbase, Voiceflow).

Why is my voice AI bill higher than the per-minute rate times my minutes?

Usually one of four rules. Partial minutes rounded up (Twilio), silence billed at the full rate (Retell) or a call left open until a silence timeout (Chatbase's default is 300 seconds), a transfer that keeps the telephony meter running after the agent stops, or a model changed from the one the quote assumed. Concurrency lines and per-minute add-ons such as recording, denoising and quality assurance are the other candidates.

Is per-minute pricing better than per-call pricing for voice AI?

It depends on your calls. Per minute rewards short, well-designed calls and charges you for rambling ones and for hold time; per call charges the same for a wrong number as for a booked appointment. The AI receptionist entry prices the same 300-call month on both meters and on a per-caller plan, and quotes each vendor's argument for its own meter.

Which voice AI platform is cheapest?

For the 1,200-minute month on this page, the lowest published total is Deepgram's agent API at $90 before the line ($60 if you bring your own model and voice, which then cost extra), but it is an API, not a product. Among managed platforms, Vapi at about $98 to $155, ElevenLabs at about $110 plus the model, and Retell at $132 to $158 are within a model choice of each other. Chatbase's credits come to about $350 at Standard for the same calls, and Intercom and Botpress do not publish a voice rate.

Related guides

• Concurrent calls — the second meter: how many lines you need, what each vendor includes, and what the caller hears when they are full.
• AI receptionist — per-minute, per-call and per-caller pricing on the same 300-call month, with the vendors' own arguments.
• Voice bot buyer's guide — the four tests to run on a vendor's demo line before signing, and how to read pricing by its shape.
• Voice and chat hybrid: when it is worth it — the decision upstream of this page.
• IVR to AI voice migration — if what you have is a phone tree, the cheapest first move is often pruning it.
• AI chatbot pricing models — the three chat pricing models this guide's voice minute sits beside.
• How to reduce chatbot costs — the chat bill's levers, priced on fourteen platforms.
• Voice bot — the pipeline this guide prices, layer by layer.

Sources

• Twilio, Programmable Voice Pricing in United States (twilio.com/en-us/voice/pricing/us), read 15 September 2026: local calls "$0.0140 / min" to make and "$0.0085 / min" to receive; toll-free "$0.0220 / min" to receive; local numbers "$1.15 / mo," toll-free "$2.15 / mo"; Conversation Relay "$0.07 / min"; Media Streams "$0.0044 / min"; call recording "$0.0025 / min"; answering machine detection "$0.0075 / call"; Gather speech recognition "$0.02 / use"; premium text-to-speech "Generative $0.0130 / 100 chars"; CPS "Provision up to 30 CPS in the console."
• Twilio Help Center, How Much Will My Voice Application Cost per Minute? (help.twilio.com/articles/223179868; the help.twilio.com page requires JavaScript, so the text was read on the support.twilio.com copy of the same article), read 15 September 2026: "Our prices are charged based on each leg of the call and any partial minutes (under 60 seconds) are rounded up to the next full minute"; the 61-second toll-free example "charged $0.044 (the 61 seconds is rounded up to 2 minutes)"; the two-leg description of a forwarded call.
• Deepgram, Pricing (deepgram.com/pricing), read 15 September 2026: "Limited-time promotional rates on streaming"; Nova-3 Monolingual streaming "Current price $0.0048/min, Regular price $0.0077/min"; Nova-3 pre-recorded "$0.0043/min"; Aura-2 "$0.030/1k characters"; Voice Agent API Standard "$0.075/min," Standard with BYO TTS "$0.065/min," Custom BYO LLM + TTS "$0.050/min," Advanced "$0.163/min" (pay-as-you-go); redaction "$0.0020/min"; keyterm prompting "$0.0013/min"; concurrency "Up to 45 for the WSS API" on the Voice Agent API.
• OpenAI, Pricing (developers.openai.com/api/docs/pricing), read 15 September 2026: "GPT-Live 1 voice sessions are billed per second, without rounding up to a whole minute. Backend model and tool usage is charged separately"; gpt-live-1 "$0.05" per minute; gpt-realtime-2.1 audio "$32.00" input and "$64.00" output per 1M tokens; gpt-live-transcribe "$0.017 / minute"; gpt-4o-transcribe "$0.006 / minute."
• Retell AI, AI Phone Agent Pricing (retellai.com/pricing), read 15 September 2026 and re-read 17 September 2026: "$0.07-$0.31 / min for AI Voice Agents"; the printed example "Cost Per Minute $0.11" with "LLM Cost $0.04/min, Retell Voice Infra $0.055/min, TTS Cost $0.015/min, Telephony Cost $0.00/min" (the page's static text; the calculator, once its script runs in a browser, preselects GPT 4.1 and shows "Cost Per Minute $0.115" with "LLM Cost $0.045/min"); the component table (Retell Voice Infra $0.055/minute; platform, Minimax, Fish, Cartesia and OpenAI voices $0.015/minute, ElevenLabs voices $0.040/minute; LLM rows as re-read 17 September 2026: GPT 6 Astra $0.32 and $0.64 fast, GPT 5.5 $0.16 and $0.32 fast, GPT 5.4 $0.080 and $0.16 fast, GPT 5.6 Terra and Claude 5 Sonnet $0.064, GPT 4.1 $0.045 and $0.0675 fast, GPT 5.1 $0.04, Claude 4.5 haiku $0.025, GPT 5.6 Luna $0.0064, GPT 4.1 nano $0.0032, GPT 5 nano $0.0016 — on 15 September the same rows read GPT 5 nano $0.003, GPT 4.1 nano $0.004 and Claude 5 Sonnet $0.08, with no GPT 6 Astra or GPT 5.6 row present; call rates $0.015/min "No Charge for Sip Trunking/Custom Telephony"; knowledge base "+$0.005/minute," batch call "+$0.005/dial," advanced denoising "+0.005/min," PII removal "+0.01/min," AI quality assurance "$0.10/min" with "First 100 minutes free"; Retell phone numbers "$2.00/month"; concurrency "Free for first 20" then "$8.00/Concurrency/month"; speech-to-speech GPT Realtime "$0.345/minute," GPT Realtime mini "$0.07/minute"); the FAQ answers on rounding ("tracked to the nearest second … No rounding up per call"), transfers ("the AI voice agent fee stops. Only the telephony fee continues"), silence ("Yes. Billing covers the entire duration of the call because the speech-to-text engine remains active and listening throughout, even during silence"), and failed calls and voicemail ("only charged for connected calls … billing applies only for the duration the AI agent is active on the line").
• Retell AI Documentation, Understand concurrency & limits (docs.retellai.com/deploy/concurrency), read 15 September 2026: burst "$0.10/min surcharge applied to the entire call duration"; "applies to the entire duration of any call that started while in burst mode"; max call duration "1 hour by default."
• Vapi, Pricing (vapi.ai/pricing), read 15 September 2026: the calculator at 1,000 minutes ("Hosting Fee 1,000 minutes $50," "Transcriber Deepgram $10," "Intelligence Model OpenAI $8 - $45," "Voice ElevenLabs $15 - $24," "Transport Fee Vapi Telephony / SIP $0," "Estimated Total $82–$129/mo") and its per-minute ranges ("$0.0095 - $0.0099/min," "$0.0077 - $0.0452/min," "$0.0146 - $0.0238/min"); "No Success Package … Pre-paid, $0.05/min Vapi hosting. Models pass-through at cost … 4 concurrent calls"; Core "$29 / month … 10 concurrent calls"; "Additional concurrency … $10 / line / month"; transport fees "Charged by provider, not Vapi": "Twilio Inbound $0.008 / min," "Twilio Outbound $0.014 / min," "Telnyx $0.0055 / min," "Vapi Telephony / SIP Free."
• ElevenLabs, ElevenAgents Pricing (elevenlabs.io/pricing/agents), read 15 September 2026: Creator "$22 … 275 minutes of calls included … 10 Concurrent Calls"; Pro "$99 … 1,238 minutes of calls included … 20 Concurrent Calls"; "Additional Call (per minute) $0.080"; "Burst pricing (per minute) $0.160"; LLM "Based on usage and varies by model"; Telephony Provider "At cost."
• ElevenLabs Help Center, How much does ElevenAgents cost? (help.elevenlabs.io/hc/en-us/articles/29298065878929), read 15 September 2026: "Voice only calls are charged based on the call duration, with a 95% discount for periods of silence longer than 10 seconds"; "The length of the call is measured based on the connection duration. This includes the time from when you begin the call, to when you end the call or the window is closed. This is why the call itself may be shorter than the duration you are charged for"; "LLM costs are passed through separately."
• Chatbase, Pricing (chatbase.co/pricing), read 15 September 2026: Standard "$150 per month … 4,000 message credits/month … Voice, Telephony"; Pro "$500 per month … 15,000 message credits/month"; "Auto recharge credits … $40 per 1000 message credits"; the comparison table's "Concurrent Calls" row (10 on Standard, 20 on Pro).
• Chatbase Documentation, Settings (chatbase.co/docs/user-guides/chatbot/settings), read 15 September 2026: "Every voice session consumes 6 message credits per minute. This is a per-minute cost regardless of the actual messages exchanged"; "Each response from the AI agent during a voice session consumes message credits based on the model used"; the pricing example table (10-minute call, 15 requests, GPT-5.6 Luna at 1 credit per request: 60 voice credits, 15 model credits, 75 total); "End conversation after silence … Defaults to 300 seconds"; "Max call duration … The default is 15 minutes."
• BotPenguin, Pricing (botpenguin.com/pricing), read 15 September 2026: King "$99 /month … AI Voice Agents — Maximum number of AI voice bots you can build — Unlimited"; Little "$29 /month" with no voice agent row; "Voice Minutes Add-ons — Extra voice minutes for AI voice agent calls — 1,000 Mins $40 /month — 5,000 Mins $150 /month."
• Voiceflow, Pricing (voiceflow.com/pricing), read 15 September 2026: "For Businesses … Book a demo, Request pricing"; "For Agencies & Partners … Transparent, usage-based billing." Voiceflow Documentation, Credits pricing table (voiceflow.com/docs/documentation/account-management/billing/credits-pricing-table, "Prices captured 2026-09-14; refreshed from the live pricing service when this page loads"), read 15 September 2026: "Phone usage — Minutes on phone calls — $0.05"; "Chat usage … $0.005"; Deepgram STT "$0.007125" and "$0.005" per minute; ElevenLabs TTS "$0.045" per 1K characters; gpt-5.6-luna "$0.00025" per 1K input tokens; gpt-5.4-mini "$0.000938."
• Intercom, Fin AI Agent Pricing (fin.ai/pricing), read 15 September 2026: "$0.99" "per outcome" with "50 outcomes per month minimum"; "Fin Voice is offered on a custom pricing plan. It's currently available to select customers working directly with our sales team."
• Botpress, Pricing (botpress.com/pricing), read 15 September 2026 by two readers: our text reader received only the page's plan-selection paragraph, which names the tiers "Pay-as-you-go," Plus, Team and Enterprise, and no rate table (the plan cards on the rendered page label the first tier Free; Botpress uses the other term in prose, as in its academy lesson How Botpress Pricing Works: "All Workspaces are on a pay-as-you-go subscription by default, which is completely free to get started with"); a second reader in the same session, using a rendered browser, saw the full page, including the comparison row marking voice for Enterprise only and the FAQ line "For voice calls, 3 minutes counts as 1 conversation." The Enterprise-only placement of the voice channel is also in our Botpress review, line 330.
• Chatbotscape review corpus (the fifteen platform reviews at /reviews), searched 15 September 2026 from the repository root. Denominator: ls sample-reviews/*-review.md | wc -l returns 15. grep -liE 'per[- ]minute' sample-reviews/*-review.md returns 2 (voiceflow, wati); grep -liE 'telephony' sample-reviews/*-review.md returns 2 (chatbase, voiceflow); grep -liE 'concurren' sample-reviews/*-review.md returns 2 (aisensy, sendpulse). Passages cited: voiceflow-review.md line 344; wati-review.md line 330; chatbase-review.md lines 167 and 220; botpress-review.md line 330.
• Ahrefs Keywords Explorer, US overview, queried 15 September 2026: the figures in this guide's keyword note.
• Chatbotscape evaluation methodology. /methodology (continuously updated).

About this guide

Chatbotscape launched in 2026 as an independent review site for chatbot platforms. This guide is part of our SMB chatbot Academy and is written for the owner or operations lead of a small business who has been quoted a per-minute rate for a voice agent and wants to know what the minute contains and which rules will make the invoice differ from the quote. It reads eleven vendors' pricing and documentation pages as published on 15 September 2026 (Retell's model-component table re-read on 17 September), takes the minute apart with their numbers, and prices one month on eight of them. It does not rank the vendors and it measured no calls.

Methodology

Every rate, limit and quotation was read on the vendor page named in Sources on 15 September 2026 and is quoted with the page's own wording; the per-minute conversions for models and voices are the platforms' own (Retell's component table, Vapi's calculator) and are not ours. Retell's component table was re-read on 17 September 2026 during the quality gate, three of its model rows had been re-priced and three more had appeared since 15 September, and both reads are printed rather than the later one silently replacing the earlier. Two pages also read differently in a text reader and in a rendered browser, and both readings are reported: Retell's calculator prints a $0.11 example in its page text and shows $0.115 with GPT 4.1 preselected once its script runs, and Botpress's pricing page gave our text reader only its plan-selection paragraph while a rendered read in the same session showed the full page. The worked month is our arithmetic on a stated scenario (300 inbound calls of four minutes on one US local number) and each line names what it leaves out; where a page did not state whether a charge exists (Chatbase and BotPenguin telephony, BotPenguin's included minutes) the line says so rather than assuming zero silently. The seven rules were chosen because each changed the worked total by more than a rounding error; the reading of what each rule means for a deployment is editorial and labelled as ours. The corpus counts reproduce with the commands printed in Sources. Nothing on this page was measured against a live call or an invoice, and the page says so wherever a reader might assume otherwise.

Last updated

17 September 2026 — Retell's model-component table re-read and every figure drawn from it corrected, with both reads printed. First published 16 September 2026.