Concurrent Calls — What the Second Meter on a Voice AI Bill Counts, How Many Lines You Need, and What Each Vendor Does When They Are Full

Quick answer: A voice bot is billed on two numbers, and the second one is the one buyers forget to ask about. Minutes are how long it talks across the month. Concurrent calls are how many callers it can hold at once, and every vendor caps that. Retell's documentation, read 15 September 2026, defines the term in one sentence: "if 15 users are engaged in voice calls with your agents at the same time, that counts as 15 concurrent calls." Vapi's documentation adds the metaphor: "Each call occupies one slot, similar to using a finite set of phone lines." The prices for those slots on the same date: Retell includes 20 per workspace and charges $8 per additional line per month; Vapi includes 4 on usage-only billing, 10 on its $29 Core package, and charges $10 per extra line; ElevenLabs includes 4 to 40 depending on the plan and, with its burst option enabled, lets calls exceed the limit up to three times over at double its per-minute rate; Chatbase gives 10 on Standard and 20 on Pro with no add-on listed. What happens to the caller who arrives when every line is busy differs more than the prices do, and that is the part of this entry worth reading before you sign.

This entry owns the term: the definition, how it relates to minutes and to calls per second, how to size it from your busiest hour, what six vendors include and charge, and what each does when the limit is hit. What a billed minute contains and the rules that change a voice bill (rounding, silence, transfers, voicemail) are the voice AI pricing guide; the per-minute versus per-call versus per-caller argument between packaged receptionist products is the AI receptionist entry.

Minutes, concurrency, and calls per second

Three quantities describe a voice agent's load, and vendors bill or cap each one separately.

The three are related but not interchangeable. A clinic that takes 300 calls a month at four minutes each buys 1,200 minutes. If those calls arrive evenly across 22 working days of eight hours, that is under two calls an hour and one line is almost always enough. If 40 of them arrive between 8:00 and 9:00 on Monday, which is how clinics work, the same 1,200 minutes need several lines for one hour a week and none the rest of the time. Minutes describe the month; concurrency describes the peak. Retell's documentation states the consequence for outbound work plainly: batch calls "are initiated as slots become available, so a batch runs no faster than your concurrency and CPS limits allow."

How many lines you need

The arithmetic is short, and it is ours, not a vendor's. Take the number of calls in your busiest hour, multiply by the average call length in minutes, and divide by 60. That is the average number of simultaneous calls during that hour. Forty calls of four minutes is 160 call-minutes in an hour, or 2.7 calls in progress at any moment on average. Calls do not arrive on average, they arrive in clumps, so the working figure is two to three times that: five to eight lines for the clinic above. On the table below that is inside Retell's 20 free lines, Vapi's 10 on its Core package, ElevenLabs' 10 on Creator and Chatbase's 10 on Standard, and above Vapi's 4 on usage-only billing and ElevenLabs' 4 on Free and 6 on Starter.

Two things change the answer. The first is call length, which is why average handle time matters here as well as on the minute bill: a bot that takes six minutes to book what should take three needs twice the lines. The second is transfers. A caller waiting for a warm transfer to a human still occupies a line until the transfer completes, and a queue of callers holding for one receptionist is a concurrency problem before it is a staffing one. Retell's dashboard has a calculator for this that asks for "calls per busy hour, average durations, and pickup rate" and returns "a recommended concurrency, inbound reservation, and CPS, each with headroom for spikes"; the inputs are the same three numbers as the arithmetic above, so the arithmetic is a reasonable first estimate before you have an account.

What six vendors include and charge

Read on 15 September 2026, monthly billing where a choice exists. Twilio is on the table because it is the telephony under several of the others, and because it caps a different quantity.

Three details in that table do more work than the prices. Vapi's usage-only allowance is four lines, which is enough for a solo practice and not for a campaign, and its four-to-ten step is the $29 Core package rather than the $10 line add-on; the same page on 3 September 2026, as recorded in our AI receptionist entry, showed a "Build" plan with 10 lines included, so the free allowance fell and the package appeared within twelve days. Retell's burst surcharge applies "to the entire duration of any call that started while in burst mode, not just the portion of time spent above the normal limit," so a ten-minute call that began one line over the limit costs an extra dollar even if the other lines cleared in the first minute. And ElevenLabs' burst is the same shape as Retell's, a switch you turn on with a ceiling of three times the plan's lines (300 at most outside Enterprise), at double the minute price; the page does not say how burst minutes are set against a plan's included minutes, so ask.

What the caller experiences when the lines are full

The prices above are for capacity you buy. The behavior below is what you get when you did not buy enough, and it is where the vendors differ most.

Rejection. Vapi's documentation says that when all slots are busy "you cannot start an outbound call or accept an inbound call until a slot becomes available," and its POST /call response carries a concurrencyBlocked flag that is "true if the call could not start because all slots were full." The documented remedy is to "Build a Twilio queue to buffer calls when you hit concurrency caps," which is a second system you configure, not a setting. Our reading: on Vapi, concurrency is a hard wall unless you build the waiting room yourself.

Wait, then fall back. Retell's documentation describes a short built-in queue: "When inbound call traffic reaches your concurrency limit, Retell briefly keeps new inbound calls waiting for an available slot. If a slot opens, the inbound call proceeds. If no slot opens after about 40 seconds," the call is transferred to the number's fallback_number if one is set, and otherwise "the call ends with concurrency_limit_reached." The fallback number is the important field: pointed at the front desk or a cell phone, it turns a busy signal into a human answering, at the telephony rate. Retell also lets a workspace hold lines for inbound callers so a campaign cannot fill them: with a limit of 100 and a reserved inbound concurrency of 20, "Outbound calls can use up to 80 slots" and "Inbound calls can use the reserved 20 slots, plus any other available slots up to the full 100."

Pay more and proceed. Retell's concurrency burst, a toggle under Settings > Limits, lets calls "that would normally be rejected for hitting your concurrency limit proceed instead, with an added surcharge" of $0.10 a minute, up to the lower of three times the limit or the limit plus 300. ElevenLabs' burst is the same idea: "When enabled, your agents can handle up to 3 times your normal concurrency limit, with excess calls charged at double the standard rate," capped at 300 calls outside Enterprise. The trade is the same on both: with the switch on, the caller never hears a busy tone, and the bill for the peak hour is higher than the bill for the rest of the day. With the switch off, Retell's documentation says the call is rejected (or, inbound, waits and falls back as above); ElevenLabs' page does not say.

Whatever the plan says. Chatbase's documentation lets you cap "Max no. of concurrent voice sessions" per agent and says the plan's default applies otherwise, and notes that the plan "enforces account-level limits for daily sessions and concurrent sessions across all your AI agents," but the page we read does not describe what the caller hears at the cap. Deepgram's pricing FAQ, collapsed in the rendered page but present in its text, says that on pay-as-you-go requests over the limit "may be queued or rejected," and points to Growth or Enterprise for custom limits; which of the two a given caller gets is not stated. For both, the honest statement is that the limit exists and the caller's experience at it is not documented where we looked; ask before the pilot, and test it by placing more calls than the plan allows.

Inbound rarely hits the cap; outbound always can

Vapi's documentation splits the two cases without hedging: inbound agents "Rarely hit concurrency caps unless traffic surges (launches, seasonal spikes)," outbound agents are "More likely to reach limits when running large calling batches." The reason is arithmetic. Inbound calls arrive at the rate customers choose to call, and a small business does not have forty customers dialing at once outside a snowstorm or a billing error. Outbound campaigns arrive at the rate the dialer chooses, which is as fast as the limit allows. Retell's FAQ confirms that batch calls "consume concurrency slots like any other outbound call," and Vapi's advice for a lead list is to "Batch long lead lists into smaller chunks (for example, 50–100 numbers) and run those batches sequentially," which is a way of saying that a 1,000-number campaign on ten lines is at most a hundred rounds of ten, each round lasting as long as the longest call in it (Retell dispatches the next number as each slot frees, which shortens the rounds but not the arithmetic of ten at a time).

The design consequence for an SMB that does both: reserve inbound lines where the platform allows it (Retell does), or run outbound campaigns outside the inbound peak, or accept that on the morning the campaign runs, the customers who call in will meet the fallback behavior above. The voice bot buyer's guide has the four tests to run on a vendor's demo line before signing; a fifth, for anyone planning campaigns, is to place more simultaneous calls than the plan includes and listen to what the extra caller hears.

What our fifteen reviews record

Searched 15 September 2026, case-insensitively, across the fifteen files matched by sample-reviews/*-review.md: concurren appears in 2 (AiSensy and SendPulse), and in neither is the word about calls. The AiSensy review uses it for how many agents share an inbox; the SendPulse review for how many client accounts an agency runs. per-minute or per minute appears in 2 (Voiceflow and Wati); the Voiceflow review records that voice deployments "require a separate Twilio account, which has its own per-minute and per-message rates," and the Wati review uses the phrase for WhatsApp message throughput. telephony appears in 2 (Voiceflow and Chatbase).

None of the fifteen protocols placed simultaneous calls against a platform's limit or recorded what a caller hears at it. Of the five platforms in our voice AI ranking, only Chatbase publishes a concurrency number on its pricing page (10 on Standard, 20 on Pro, read 15 September 2026); Voiceflow's business pricing is demo-gated, BotPenguin's pricing page sells voice minutes in packs (1,000 for $40 a month, 5,000 for $150) and does not state a line count, Intercom's Fin Voice is "offered on a custom pricing plan," and Botpress gates voice to its Enterprise tier. That is a gap in the protocol and in the market's disclosure, recorded here, not a finding about any platform. Corrections to editorial@chatbotscape.com.

Related terms

• Voice bot — the category this capacity applies to, and why voice is harder than chat.
• AI receptionist — the per-minute, per-call and per-caller meters compared on the same 300-call month.
• Average handle time — the call length that, multiplied by busy-hour volume, sets the lines you need.
• Warm transfer — the hand-off during which the caller still occupies a line.
• Interactive voice response — the phone tree that concurrency limits used to belong to.
• Speech-to-text — the layer that stays open, and billed, for the whole call.

FAQ

What does "concurrent calls" mean on a voice AI pricing page?

The number of phone conversations the agent can hold at the same instant. Retell's documentation gives the example of 15 users on calls at once counting as 15 concurrent calls. It is a cap on capacity, separate from the minutes you are billed for, and most vendors include a number of lines with the plan and sell more by the month.

How many concurrent calls does a small business need?

Take the calls in your busiest hour, multiply by the average call length in minutes, divide by 60, then double or triple it for clumping. Forty four-minute calls in the busiest hour is 2.7 simultaneous calls on average and five to eight lines with headroom. That is inside Retell's 20 free lines, Vapi's 10 on its Core package, ElevenLabs' 10 on Creator, and Chatbase's 10 on Standard, and above Vapi's 4 on usage-only billing.

What happens to a caller when all the lines are busy?

It depends on the vendor. On Vapi the call cannot start until a slot frees, and a queue is something you build. On Retell an inbound call waits about 40 seconds, then transfers to a fallback number if you set one, or ends. With burst enabled, on Retell or on ElevenLabs, the call proceeds up to three times the limit and is billed at a surcharge ($0.10 a minute on Retell) or at double the minute rate ($0.16 on ElevenLabs). Chatbase documents the limit but not the caller's experience at it; Deepgram's FAQ says requests over the limit "may be queued or rejected."

Does an outbound campaign use the same lines as inbound calls?

Yes. Retell's FAQ says all active voice calls "draw from the same concurrency pool" and that batch calls consume slots like any other outbound call; Retell lets you reserve a share for inbound. Vapi's advice is to run lead lists in batches of 50 to 100 numbers sequentially. Without a reservation, a campaign running during your inbound peak means inbound callers meet the full-lines behavior.

Is concurrency the same as calls per second?

No. Concurrency is how many calls are open at once; calls per second is how fast new ones can start. Twilio's pricing page meters CPS (up to 30 provisionable in its console, 1 free on pay-as-you-go) and does not mention a concurrency cap; Retell sets CPS per telephony path and notes that custom-telephony CPS "scales with your concurrency." A campaign is limited by both: it can dial no faster than CPS allows and hold no more calls than concurrency allows.

Sources

• Vapi, Pricing (vapi.ai/pricing), read 15 September 2026: "No Success Package … Pre-paid, $0.05/min Vapi hosting. Models pass-through at cost … 4 concurrent calls … 1 included Vapi phone number … 14 days raw data retention"; Core "$29 / month … 10 concurrent calls … 5 included Vapi phone numbers"; Pro "10% of Vapi hosting fee, $999/mo minimum … 30 concurrent calls"; Premier "Elevated orgs, concurrency, and data retention"; add-on "Additional concurrency — Extra concurrent call lines beyond your package default — $10 / line / month"; "Additional concurrency and organizations can be added as add-ons at any package."
• Vapi Documentation, Understanding Call Concurrency (docs.vapi.ai/calls/call-concurrency), read 15 September 2026: "Call concurrency represents how many Vapi calls can be active at the same time. Each call occupies one slot, similar to using a finite set of phone lines"; "When all slots are busy, you cannot start an outbound call or accept an inbound call until a slot becomes available"; inbound agents "Rarely hit concurrency caps unless traffic surges (launches, seasonal spikes)"; outbound agents "More likely to reach limits when running large calling batches"; "Batch long lead lists into smaller chunks (for example, 50–100 numbers) and run those batches sequentially"; the subscriptionLimits fields concurrencyBlocked ("true if the call could not start because all slots were full"), concurrencyLimit and remainingConcurrentCalls; "Build a Twilio queue to buffer calls when you hit concurrency caps."
• Retell AI, AI Phone Agent Pricing (retellai.com/pricing), read 15 September 2026: "20 Free Concurrent Calls"; "Concurrent calls — 20 included, more available on demand" (pay-as-you-go) and "No Cap on Concurrent Calls" (Enterprise, in the comparison table; the Enterprise plan card on the same page says "High Cap on Concurrent Calls"); "Concurrency (Active Calls) — Free for first 20 concurrency (active calls) — $8.00/Concurrency/month"; FAQ "Every account includes 20 concurrent calls for free. Need more? Additional capacity is $8 per concurrent call per month. Add or remove anytime as your volume changes. Enterprise plans offer custom concurrency starting at 50+ calls."
• Retell AI Documentation, Understand concurrency & limits (docs.retellai.com/deploy/concurrency), read 15 September 2026: the definition ("if 15 users are engaged in voice calls with your agents at the same time, that counts as 15 concurrent calls"); "Concurrency limits apply per workspace, not per account"; "Pay-As-You-Go workspaces are allocated a quota of 20 concurrent calls by default"; "Each agent can handle an unlimited number of calls, as long as total concurrency stays within your quota"; the reserved-inbound example (limit 100, reserved 20: "Outbound calls can use up to 80 slots"); "If no slot opens after about 40 seconds" with the fallback_number transfer and the concurrency_limit_reached end reason; burst "$0.10/min surcharge applied to the entire call duration," the burst limit as "the lower of 3× your concurrency limit, OR your concurrency limit + 300," and the note that the surcharge "applies to the entire duration of any call that started while in burst mode"; the calculator inputs ("calls per busy hour, average durations, and pickup rate") and outputs; "Custom Telephony CPS scales with your concurrency"; max call duration "1 hour by default … up to 2 hours"; FAQ answers on the shared pool and on batch calls ("a batch runs no faster than your concurrency and CPS limits allow").
• ElevenLabs, ElevenAgents Pricing (elevenlabs.io/pricing/agents), read 15 September 2026: Free "15 minutes of calls included, 4 Concurrent Calls"; Starter "$6 … 75 minutes … 6 Concurrent Calls"; Creator "$22 … 275 minutes … 10 Concurrent Calls"; Pro "$99 … 1,238 minutes … 20 Concurrent Calls"; Scale "$299 … 3,738 minutes … 30 Concurrent Calls"; Business "$990 … 12,375 minutes … 40 Concurrent Calls"; Enterprise "Elevated concurrency limits"; the pricing table rows "Additional Call (per minute) $0.080," "Text message (per message) $0.003," "Burst pricing (per minute) $0.160"; LLM "Based on usage and varies by model," telephony provider "At cost"; the FAQ "What is burst pricing?" (collapsed in the rendered page, present in its text): "Burst pricing allows your ElevenLabs agents to temporarily exceed your workspace's subscription concurrency limit during high-demand periods. When enabled, your agents can handle up to 3 times your normal concurrency limit, with excess calls charged at double the standard rate," and the Burst pricing (per minute) row's tooltip, "with a maximum of 300 calls for non-enterprise customers."
• 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"; "LLM costs are passed through separately."
• Deepgram, Pricing (deepgram.com/pricing), read 15 September 2026: "Usage & Scale (Concurrency)" rows for Speech-to-Text ("Up to 50 for the REST API," "Up to 150 for the WSS API" on Pay As You Go and "Up to 225" on Growth), Text-to-Speech ("Up to 45" and "Up to 60") and Voice Agent API ("Up to 45 for the WSS API" and "Up to 60"); Voice Agent API Standard "$0.075/min" pay-as-you-go; the FAQ entry "What happens if I exceed my concurrency limit?" (collapsed in the rendered page, present in its text): on the Pay-As-You-Go plan, requests exceeding the limit "may be queued or rejected. To avoid this, you can upgrade to a Growth or Enterprise plan, which offers custom concurrency limits."
• BotPenguin, Pricing (botpenguin.com/pricing), read 15 September 2026: "Voice Minutes Add-ons — Extra voice minutes for AI voice agent calls — 1,000 Mins $40 /month — 5,000 Mins $150 /month"; no concurrency figure on the page.
• 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"; the comparison table's "Concurrent Calls" row (10 on Standard, 20 on Pro; no voice on Free or Hobby); Enterprise "Higher limits."
• Chatbase Documentation, Settings (chatbase.co/docs/user-guides/chatbot/settings), read 15 September 2026: "Max no. of concurrent voice sessions: The maximum number of voice sessions that can run at the same time. If left empty, your plan's default limit will apply"; "Your subscription plan enforces account-level limits for daily sessions and concurrent sessions across all your AI agents"; "Every voice session consumes 6 message credits per minute."
• Twilio, Programmable Voice Pricing in United States (twilio.com/en-us/voice/pricing/us), read 15 September 2026: "Calls per second (CPS) pricing … Provision up to 30 CPS in the console"; the CPS slider showing 1 CPS at "Pay as you go Pricing — Free"; local numbers "$0.0085 / min + $1.15 / mo" to receive calls.
• Intercom, Fin AI Agent Pricing (fin.ai/pricing), read 15 September 2026: "Fin Voice is offered on a custom pricing plan. It's currently available to select customers working directly with our sales team."
• Chatbotscape review corpus (the fifteen platform reviews listed at /reviews), searched 15 September 2026 from the repository root. Denominator: ls sample-reviews/*-review.md | wc -l returns 15. grep -liE 'concurren' sample-reviews/*-review.md returns 2 (aisensy, sendpulse); grep -liE 'per[- ]minute' sample-reviews/*-review.md returns 2 (voiceflow, wati); grep -liE 'telephony' sample-reviews/*-review.md returns 2 (chatbase, voiceflow). Passages cited: aisensy-review.md line 185; sendpulse-review.md line 175; voiceflow-review.md line 344; wati-review.md line 330; botpress-review.md line 330.
• Ahrefs Keywords Explorer, US overview, queried 15 September 2026 — the demand, difficulty and parent-topic figures in this entry's keyword note.
• Chatbotscape evaluation methodology. /methodology (continuously updated).